Methods, devices, equipment, media, and programs for generating phishing practice emails.

This method for generating structured constraints by collecting intranet access events and organizing neighborhood email corpora, and then performing multi-dimensional verification of phishing drill emails, solves the problems of business disconnect and privacy leakage in existing technologies for phishing email generation, and achieves business-related and secure phishing drill email generation.

CN123122870APending Publication Date: 2026-11-17CHINA UNITED NETWORK COMM GRP CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611097203.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-22
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing phishing email generation technologies struggle to simultaneously ensure the relevance of training emails to business needs, privacy and security, and controllable similarity, leading to issues such as templates being disconnected from business requirements, high risks of privacy leaks, and high costs of manual review.

Method used

By collecting intranet access events of target personnel and pre-set evaluation indicators, business scenario characteristic information is determined. Combined with email corpus from the organization's neighboring domain, email corpus characteristic information is extracted to generate structured target constraints. Multi-dimensional verification is then performed to generate phishing drill emails, ensuring that the emails are within the business boundaries and conform to organizational habits.

Benefits of technology

Without copying real emails or disclosing sensitive information, this method generates phishing practice emails that are relevant to the target personnel's recent business and whose expression conforms to the organization's habits. This solves the problems of templates being out of touch with business, high risks of privacy leaks, and high costs of manual review, and achieves secure and compliant generation of phishing practice emails.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123122870A_ABST
    Figure CN123122870A_ABST
Patent Text Reader

Abstract

This application provides a method, apparatus, device, medium, and program product for generating phishing drill emails, relating to the field of network security technology, to improve the ability of phishing drill emails to balance business relevance, privacy security, and controllable similarity. The specific technical solution is as follows: Business scenario feature information of the initial drill email is determined based on the target personnel's internal network access events and preset evaluation indicators; email corpus feature information is generated based on email corpus from the target personnel's organizational neighborhood; target constraints are generated based on the business scenario feature information and email corpus feature information; the initial drill email is generated based on the target constraints; multi-dimensional verification is performed on the initial drill email, and the verification result is output; if the verification result indicates that the initial drill email passes the multi-dimensional verification, the initial drill email is identified as the target phishing drill email. This application is applied in internal network security drill scenarios.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of network security technology, and in particular to a method, apparatus, device, medium, and program product for phishing drill emails. Background Technology

[0002] With the development of network attack and defense technologies, phishing email attacks have become one of the main threats to enterprise internal network security. To improve employees' ability to identify suspicious emails, security teams typically conduct simulation tests with employees through phishing email drills.

[0003] In related technologies, phishing drill email generation methods mainly fall into the following categories: First, generating emails using fixed templates. These templates have weak relevance to employees' actual business and are easily identified. Second, generating emails using large models. However, large models are prone to fabricating non-existent business systems or introducing sensitive words, requiring extensive manual review. Third, rewriting generated emails by analyzing historical emails within the company. However, historical emails contain sensitive information such as customer names and contract numbers, posing a risk of privacy leaks, and the similarity between the generated content and real emails is uncontrollable.

[0004] Therefore, the relevant technologies all struggle to simultaneously ensure the business relevance, privacy and security, and controllability of similarity in training emails. Summary of the Invention

[0005] This application provides a method, apparatus, device, medium, and program product for generating phishing practice emails, which can improve the ability of phishing practice emails to balance business relevance, privacy security, and controllable similarity.

[0006] In a first aspect, embodiments of this application provide a method for generating phishing drill emails, the method comprising:

[0007] The business scenario feature information of the initial exercise email is determined based on the intranet access events of the target personnel and the preset evaluation indicators, and the email corpus feature information is generated based on the email corpus of the target personnel's organizational neighborhood.

[0008] Based on the business scenario feature information and the email corpus feature information, target constraints are generated. The target constraints include at least a list of allowed fields, a list of prohibited content rules, and writing constraints.

[0009] Generate initial exercise emails based on the aforementioned target constraints;

[0010] Perform multi-dimensional verification on the initial exercise emails and output the verification results. The multi-dimensional verification includes at least real email similarity verification and business boundary and consistency verification.

[0011] If the verification result indicates that the initial training email passes the multi-dimensional verification, the initial training email is identified as a target phishing training email.

[0012] The technical solution provided in this application brings at least the following beneficial effects:

[0013] The method in this application determines business scenario feature information by collecting intranet access events of target personnel and combining them with preset evaluation indicators. Simultaneously, it extracts email corpus feature information from email corpora within the target personnel's organizational network. Thus, before email generation, it both identifies email topics highly relevant to the target personnel's recent business activities and understands the expression habits and structural patterns of real emails within the personnel's organizational environment. Based on this, the above two types of feature information are transformed into structured target constraints, thereby limiting the subsequent email generation process within preset business boundaries, content boundaries, and expression boundaries. This avoids the privacy leakage risks associated with directly using the original email body or real sensitive information, and also prevents... This method addresses issues such as fabricated business systems, process errors, or the introduction of sensitive words. Therefore, under the constraints of the target conditions, it generates an initial training email and performs multi-dimensional checks on this initial training email, including at least real email similarity checks and business boundary and consistency checks. Real email similarity checks ensure that the generated email will not lose its training significance or cause privacy compliance risks due to excessive similarity to real emails. Business boundary and consistency checks ensure that the generated email does not exceed the allowed field range and is consistent with the actual business scenario of the target personnel. Only when the verification results indicate that the initial training email passes all checks is it identified as a target phishing training email, thus forming a final line of security and compliance defense after the email is generated. In summary, the method of this embodiment achieves the generation of phishing training emails that are relevant to the target personnel's recent business and whose expression conforms to organizational habits without copying real emails, disclosing sensitive information, or exceeding business boundaries. This effectively solves the problems of templates being disconnected from business, high privacy leakage risks, large-scale fabrication of business systems, and high manual review costs in existing technologies.

[0014] One possible implementation is that the preset evaluation indicators include access frequency, recent access time, system sensitivity level, and operation type;

[0015] The business scenario characteristic information for determining the initial exercise email based on the target personnel's intranet access events and preset evaluation indicators includes:

[0016] Based on the intranet access events, determine at least one business system accessed by the target personnel;

[0017] The popularity value of each business system is determined according to the preset evaluation indicators;

[0018] The business scenario feature information is determined based on the business system with the highest popularity value.

[0019] Another possible implementation is that the email corpus feature information includes topic structure distribution, body structure distribution, tone style distribution, business action distribution, and body length distribution;

[0020] The step of generating email corpus feature information based on email corpus from the target person's organizational neighborhood includes:

[0021] Sensitive entity identification and de-identification processing are performed on the email corpus, and feature information of the email corpus is extracted.

[0022] Another possible implementation is that the target constraint also includes displaying the sender's identity;

[0023] Before generating target constraints based on the business scenario feature information and the email corpus feature information, the phishing exercise email generation method further includes:

[0024] Obtain the email communication metadata of the target personnel, wherein the email communication metadata includes at least one of the following: sender, recipient, sending time, sender identity type, email subject category, and associated business system;

[0025] Based on the email communication metadata, a communication relationship diagram centered on the target person is constructed, and the communication relationship diagram includes the communication relationship between the target person and at least one candidate sender identity;

[0026] The overall matching degree of each candidate sender identity is determined based on the communication relationship diagram, and the sender identity to be displayed is determined from each candidate sender identity based on the overall matching degree.

[0027] Another possible implementation is that the multidimensional verification also includes writing structure verification and sensitive information verification;

[0028] The process of performing multi-dimensional verification on the initial training email and outputting the verification results includes:

[0029] The initial training emails are subjected to the following checks: similarity to real emails, business boundary and consistency, writing structure and sensitive information, generating multiple single-dimensional check results.

[0030] If all the single-dimensional verification results indicate that the initial training email has passed the multi-dimensional verification, then the initial training email is determined to have passed the multi-dimensional verification; or

[0031] If any single-dimensional verification result fails, the initial training email is determined to have failed the multi-dimensional verification.

[0032] Another possible implementation is to mark the initial training email as failing the multidimensional verification if the verification result indicates that the initial training email has failed the verification, and to regenerate candidate training emails based on the target constraints.

[0033] Secondly, embodiments of this application provide a phishing drill email generation device, comprising:

[0034] The determination module is used to determine the business scenario feature information of the initial exercise email based on the intranet access events of the target personnel and preset evaluation indicators, and to generate email corpus feature information based on the email corpus of the target personnel's organizational neighborhood.

[0035] The generation module is used to generate target constraints based on the business scenario feature information and the email corpus feature information. The target constraints include at least a list of allowed fields, a list of prohibited content rules, and writing constraints.

[0036] The generation module is also used to generate initial exercise emails based on the target constraints.

[0037] The verification module is used to perform multi-dimensional verification on the initial exercise email and output the verification results. The multi-dimensional verification includes at least real email similarity verification and business boundary and consistency verification.

[0038] The determining module is further configured to determine the initial training email as a target phishing training email when the verification result indicates that the initial training email has passed the multi-dimensional verification.

[0039] One possible implementation is that the preset evaluation indicators include access frequency, recent access time, system sensitivity level, and operation type;

[0040] The determining module is specifically used for:

[0041] Based on the intranet access events, determine at least one business system accessed by the target personnel;

[0042] The popularity value of each business system is determined according to the preset evaluation indicators;

[0043] The business scenario feature information is determined based on the business system with the highest popularity value.

[0044] Another possible implementation is that the email corpus feature information includes topic structure distribution, body structure distribution, tone style distribution, business action distribution, and body length distribution;

[0045] The determining module is specifically used for:

[0046] Sensitive entity identification and de-identification processing are performed on the email corpus, and feature information of the email corpus is extracted.

[0047] Another possible implementation is that the target constraint also includes displaying the sender's identity;

[0048] The generation module is further configured to:

[0049] Obtain the email communication metadata of the target personnel, wherein the email communication metadata includes at least one of the following: sender, recipient, sending time, sender identity type, email subject category, and associated business system;

[0050] Based on the email communication metadata, a communication relationship diagram centered on the target person is constructed, and the communication relationship diagram includes the communication relationship between the target person and at least one candidate sender identity;

[0051] The overall matching degree of each candidate sender identity is determined based on the communication relationship diagram, and the sender identity to be displayed is determined from each candidate sender identity based on the overall matching degree.

[0052] Another possible implementation is that the multidimensional verification also includes writing structure verification and sensitive information verification;

[0053] The verification module is specifically used for:

[0054] The initial training emails are subjected to the following checks: similarity to real emails, business boundary and consistency, writing structure and sensitive information, generating multiple single-dimensional check results.

[0055] If all the single-dimensional verification results indicate that the initial training email has passed the multi-dimensional verification, then the initial training email is determined to have passed the multi-dimensional verification; or

[0056] If any single-dimensional verification result fails, the initial training email is determined to have failed the multi-dimensional verification.

[0057] In another possible implementation, the verification module is further used for:

[0058] If the verification result indicates that the initial training email has failed the multidimensional verification, the initial training email is marked as verification failed, and candidate training emails are regenerated based on the target constraints.

[0059] Thirdly, this application provides an electronic device comprising: a processor and a memory; the memory stores a program or instructions executable on the processor, wherein the program or instructions, when executed by the processor, implement the method of the first aspect described above.

[0060] Fourthly, this application provides a readable storage medium on which a program or instructions are stored, which, when executed by a computer, implement the method of the first aspect described above.

[0061] Fifthly, this application provides a computer program product stored in a storage medium, which, when executed by a computer, implements the method described in the first aspect.

[0062] In a sixth aspect, embodiments of this application provide a chip including a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run programs or instructions to implement the method described in the first aspect.

[0063] The beneficial effects of the second to sixth aspects mentioned above are described in the corresponding description of the first aspect and will not be repeated here. Attached Figure Description

[0064] Figure 1 The network architecture for a phishing drill email generation method provided in this application embodiment.

[0065] Figure 2 This is a flowchart illustrating a method for generating phishing practice emails, as provided in an embodiment of this application.

[0066] Figure 3 This is a flowchart illustrating a method for determining business scenario feature information provided in an embodiment of this application.

[0067] Figure 4 This is a flowchart illustrating a method for generating email corpus feature information according to an embodiment of this application.

[0068] Figure 5 This is a flowchart illustrating a method for determining the sender's identity, provided as an embodiment of this application.

[0069] Figure 6 This is a flowchart illustrating a multidimensional verification method provided in an embodiment of this application.

[0070] Figure 7 This is a schematic diagram of the overall architecture of a phishing drill email generation system provided in an embodiment of this application.

[0071] Figure 8 This is a schematic diagram of the overall process of a phishing drill email generation method provided in an embodiment of this application.

[0072] Figure 9 This is a flowchart illustrating a process for determining business scenario feature information based on intranet access events of target personnel and preset evaluation indicators, as provided in an embodiment of this application.

[0073] Figure 10 This is a schematic diagram illustrating a process for constructing a communication relationship diagram based on email communication metadata, determining the comprehensive matching degree of each candidate sender's identity, and determining the sender's identity to be displayed, as provided in an embodiment of this application.

[0074] Figure 11 This is a flowchart illustrating a process for sensitive entity identification and de-identification of email corpora from the organizational neighborhood of a target person, and for extracting feature information from the email corpora, as provided in this application embodiment.

[0075] Figure 12 This is a schematic diagram of a phishing practice email generation device provided in an embodiment of this application.

[0076] Figure 13 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0077] The following is a detailed description of the phishing exercise email generation method, apparatus, equipment, medium, and program products provided in this application, with reference to the accompanying drawings.

[0078] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0079] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0080] The terms "at least one," "at least one of," etc., used in the specification and claims of this application refer to any one, any two, or a combination of two or more of the included items. For example, at least one of a, b, and c can mean: "a," "b," "c," "a and b," "a and c," "b and c," and "a, b, and c," where a, b, and c can be single or multiple. Similarly, "at least two" refers to two or more items, and its meaning is similar to that of "at least one."

[0081] In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0082] The embodiments of this application provide a method, apparatus, device, medium, and program product for generating phishing drill emails, which can be applied to internal network security drill scenarios.

[0083] Among related technologies, phishing drill email generation methods mainly fall into three categories: those based on fixed templates, those based on large models, and those based on historical email analysis. While fixed template methods are simple to implement, the content has weak relevance to employees' actual business, making it easily identifiable. Large model methods generate email text by engineering prompts on target information; however, without effective constraints, the model can easily fabricate non-existent business systems or introduce sensitive words, requiring extensive manual review of the generated results. Historical email analysis methods extract and rewrite the semantic style of real emails, but the original emails often contain a large amount of sensitive information such as customer names and contract numbers; direct use poses privacy and compliance risks, and the similarity between the generated content and real emails is uncontrollable. In summary, none of these technologies can simultaneously achieve a balance between business relevance, privacy security, and controllable similarity in drill emails.

[0084] To address the aforementioned technical problems, this application provides a method, apparatus, device, medium, and program product for generating phishing practice emails. This method determines business scenario characteristic information by collecting intranet access events of target personnel and combining them with preset evaluation indicators. Simultaneously, it extracts email corpus characteristic information from email corpora within the target personnel's organizational network. Thus, before email generation, it not only identifies email topics highly relevant to the target personnel's recent actual business behavior but also grasps the expression habits and structural patterns of real emails within the personnel's organizational environment. Based on this, the above two types of characteristic information are transformed into structured target constraints, thereby limiting the subsequent email generation process within preset business boundaries, content boundaries, and expression boundaries, avoiding the direct use of the original email body or truly sensitive content. This method addresses the privacy risks associated with information leakage while also preventing issues such as fabricated business systems, process errors, or the introduction of sensitive words. Furthermore, under the constraints of the target conditions, this method generates an initial training email and performs multi-dimensional checks on it, including at least real email similarity checks and business boundary and consistency checks. Real email similarity checks ensure that the generated email will not lose its training significance or cause privacy compliance risks due to excessive similarity to real emails. Business boundary and consistency checks ensure that the generated email does not exceed the allowed field range and is consistent with the target personnel's actual business scenario. Only when the verification results indicate that the initial training email passes all checks is it identified as a target phishing training email, thus forming a final line of security and compliance defense after the email is generated. In summary, the method of this embodiment achieves the generation of phishing training emails that are relevant to the target personnel's recent business and whose expression conforms to organizational habits without copying real emails, leaking sensitive information, or exceeding business boundaries. This effectively solves the problems of templates being disconnected from business, high privacy leakage risks, large-scale fabrication of business systems, and high costs of manual review in existing technologies.

[0085] The following description, in conjunction with the accompanying drawings, details the phishing exercise email generation device, equipment, medium, and program products provided in the embodiments of this application.

[0086] Figure 1 The network architecture of a phishing drill email generation method provided in an embodiment of this application is illustrated. For example... Figure 1 As shown, the network architecture includes a phishing practice email generation device 101 and a terminal device 102. The phishing practice email generation device 101 and the terminal device 102 are interconnected.

[0087] In some embodiments, the phishing practice email generation device 101 may be a server, a computer, or a processor or processing unit within a server or computer. The server may be a single server or a server cluster consisting of multiple servers. It should be noted that this application embodiment does not limit the specific device form of the phishing practice email generation device 101. Figure 1 The example shown is a phishing drill email generation device 101, which is a single server.

[0088] In some embodiments, the terminal device may be a mobile phone, tablet computer, laptop computer, handheld computer, in-vehicle electronic device, mobile internet device (MID), augmented reality (AR) / virtual reality (VR) device, robot, wearable device, personal computer (PC), ultra-mobile personal computer (UMPC), netbook, or personal digital assistant (PDA), etc., and the embodiments of this application do not specifically limit it. Figure 1 The example shown is a mobile phone, with terminal device 102 as an example.

[0089] In some embodiments, security managers or exercise task configuration personnel can log in to the management platform or operation interface provided by the phishing exercise email generation device 101 through terminal device 102, send exercise task configuration information and a list of target exercise personnel to the phishing exercise email generation device 101, and view the generated candidate exercise emails, verification results, audit records, and the final target phishing exercise email. Correspondingly, after generating the target phishing exercise email, the phishing exercise email generation device 101 can send the target phishing exercise email to the terminal device 102 through the network. The terminal device 102 then displays the content, verification details, and audit records of the target phishing exercise email to the security managers in the form of an email client, a security exercise platform interface, or a notification message, so that the security managers can confirm, review, or manually intervene in the generation results. In some embodiments, the terminal device 102 is further configured to receive a confirmation instruction returned by the phishing exercise email generation device 101 after the target phishing exercise email has been confirmed by the security management personnel, and send the target phishing exercise email to the receiving terminal corresponding to the target exerciser, or send the target phishing exercise email back to the phishing exercise email generation device 101, which then delivers the target phishing exercise email to the target exerciser through an email gateway or email sending module.

[0090] It should be noted that the network architecture described in the embodiments of this application is for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and does not constitute a limitation on the technical solutions provided in the embodiments of this application. As network architectures evolve, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0091] See Figure 2 This is a flowchart illustrating a method for generating phishing practice emails according to an embodiment of this application. Figure 2 As shown, the phishing exercise email generation method provided in this application embodiment can be implemented by the above-mentioned phishing exercise email generation device, specifically including the following steps 201 to 205.

[0092] Step 201: The phishing exercise email generation device determines the business scenario feature information of the initial exercise email based on the target personnel's intranet access events and preset evaluation indicators, and generates email corpus feature information based on the email corpus of the target personnel's organizational neighborhood.

[0093] Step 202: The phishing exercise email generation device generates target constraints based on business scenario feature information and email corpus feature information. The target constraints include at least a list of allowed fields, a list of prohibited content rules, and writing constraints.

[0094] Step 203: The phishing exercise email generation device generates an initial exercise email based on the target constraints.

[0095] Step 204: The phishing exercise email generation device performs multi-dimensional verification on the initial exercise email and outputs the verification results. The multi-dimensional verification includes at least real email similarity verification and business boundary and consistency verification.

[0096] Step 205: If the verification result indicates that the initial training email has passed the multi-dimensional verification, the phishing training email generation device will identify the initial training email as the target phishing training email.

[0097] First, the business scenario characteristics of the initial exercise emails are determined based on the intranet access events of the target personnel and the preset evaluation indicators, and the email corpus characteristics are generated based on the email corpus of the target personnel's organizational neighborhood.

[0098] In this embodiment, the target personnel include internal employees of the enterprise or third-party personnel who need to conduct phishing email drills. They can be specified by the drill task list or automatically determined by the system according to risk grouping or statistical sampling methods. The identification information of the target personnel can be, for example, a corporate email account or employee ID.

[0099] Internal network access events include access records generated when target personnel access various business systems within the enterprise's internal network environment. It is understood that internal network access events can be obtained from data sources such as the unified authentication system, VPN gateway logs, zero-trust gateway audit records, bastion host operation logs, business system's own audit logs, or API gateway logs. Specifically, each internal network access event can include the following information: user identifier, accessed business system identifier and name, system type, access time, operation type, access result, access path, and system sensitivity level. This information collectively reflects which operations the target personnel recently performed on which business systems, thus providing factual basis for determining the business themes to be addressed in subsequent exercise emails.

[0100] The pre-defined evaluation metrics include a set of standards used to identify the business systems most relevant to the target personnel from intranet access events. Pre-defined evaluation metrics may include at least one of the following: access frequency (i.e., the number of times the target personnel access a particular business system within a statistical time window), recent access time (i.e., the time interval between the target personnel's most recent access to a particular business system and the current time), system sensitivity level (i.e., the sensitivity of the business system in the enterprise security classification), and operation type (i.e., the specific type of operation performed by the target personnel on the business system, such as submit, approve, download, etc.).

[0101] Business scenario characteristic information includes information describing the characteristics of the business topic, business system, process stage, and business actions that the exercise email should revolve around. For example, business scenario characteristic information may include: the name and type of the target business system, the process stage that the target personnel are in within that business system (e.g., "after permission renewal submission / before approval confirmation"), the set of candidate business actions associated with that process stage (e.g., "permission review reminder", "status confirmation reminder", "supplementary information reminder"), and the suggested email subject type, etc.

[0102] Organizational neighborhood encompasses the set of individuals who are related to the target person in terms of organizational structure, business collaboration, or job function. There are various ways to determine organizational neighborhood, such as based on at least one of the target person's department, project team, job position, email groups, or business system user groups. Organizational neighborhood limits the source range of email corpora, ensuring that the extracted email expression characteristics reflect the actual communication habits of the target person's business team.

[0103] The email corpus comprises email samples sent and received by personnel within the organization's neighborhood during the statistical time period. It's important to note that in this step, the original emails are only used to extract email corpus feature information and are not directly provided to the email generation model in subsequent steps. Specifically, the system first performs sensitive entity identification and de-identification on the email corpus. Sensitive entities may include, for example, real personnel names, customer names, project codes, contract numbers, order numbers, amounts, mobile phone numbers, ID card numbers, email accounts, internal IP addresses, real domain names, real attachment names, and access credentials. Then, the de-identified text or structured intermediate results are input into the corpus feature extraction component to extract email corpus feature information. This email corpus feature information includes abstract features extracted from the organization's neighborhood email corpus after de-identification and statistical analysis. Examples include topic structure distribution, body text structure distribution, tone style distribution, business action distribution, and body text length distribution. This feature information reflects the expression habits and style preferences of personnel within the organization's neighborhood in writing emails, providing a reference for generating writing constraints in the subsequent target constraints.

[0104] Then, target constraints are generated based on business scenario features and email corpus features.

[0105] In this embodiment, the business scenario feature information and email corpus feature information obtained in the preceding steps are integrated into structured target constraints. These target constraints are data structures used to constrain the input range and output boundaries of a large language model. Their function is to ensure that the large model, when generating emails, does not exceed the preset business scope, does not generate sensitive content, and conforms to the email expression habits of the target person's organization.

[0106] In this embodiment, the target constraints include at least a list of allowed fields, a list of prohibited content rules, and writing constraints.

[0107] The allowed field list is used to limit the business systems, business actions, process stages, and subject types that the large language model can use when generating emails. For example, if the current business scenario is "permission renewal for the permission management platform," the allowed field list can include the business system "permission management platform," the business actions "permission review reminder" and "status confirmation reminder," the process stage "permission renewal submitted / approved before confirmation," and the subject type "business processing notification." The allowed field list can be generated based on business scenario characteristics, combined with the statistical results of a pre-set business thesaurus, business action mapping table, and anonymized organizational neighborhood corpus. For example, the large language model can be used to categorize or sort candidate fields, but the large language model will not add business systems, business actions, or subject types that do not appear in the business thesaurus, business action mapping table, or pre-analysis results.

[0108] The prohibited content rule list is used to constrain the types and formats of content that the large language model should avoid generating, preventing the model from fabricating sensitive formats or information. The prohibited content rule list records the types of content that must not be generated, such as: customer name entities, contract number entities, application form number entities, account number entities, password / credential entity entities, actual attachment name entities, and amount entities. It should be noted that the prohibited content rule list only records the types and formats of content that must not be generated, and does not record specific sensitive values ​​such as actual customer names and actual document numbers identified from the original email. These actual sensitive values ​​will not be included in the target constraints after the aforementioned anonymization process, thus avoiding the risk of privacy leakage at the source.

[0109] Writing constraints are used to limit the structure, tone, length, and expression of candidate emails. These constraints are generated based on statistical results from email corpus features, such as the distribution of topic structure, body structure, tone style, business actions, and body length. For example, if emails within an organization related to this business scenario typically use a body structure of "salutation + identification + description + instructions + contact information," a formal yet relaxed tone, and a body length of 200-400 words, then writing constraints can be applied to impose requirements on the structure, tone, and length of candidate emails.

[0110] After generating the target constraints, an initial exercise email is generated based on the target constraints.

[0111] In this embodiment, the target constraints are used as input to the large language model, which then generates an initial training email. Since the target constraints have clearly defined the scope of the large language model's generation through a list of allowed fields, a list of prohibited content rules, and writing constraints, the large language model can only generate email content related to the target personnel's recent business scenarios. It will not automatically expand to unrelated business topics, nor can it generate sensitive entities or real business numbers. Furthermore, the information received by the large language model in this step does not include original access logs, original real email text, or any un-anonymized sensitive data.

[0112] The initial training emails output by the large language model can be encapsulated into candidate email objects. A candidate email object can include at least: candidate email ID, email subject, displayed sender name, displayed sender email address, email body, training entry placeholder, business action used, and structured context number. The training entry placeholder is a special marker pre-reserved in the email text, used in subsequent steps to replace the original with a unique link or identifier generated by the training platform, rather than directly generating a real training link in this step.

[0113] Next, multi-dimensional verification is performed on the initial exercise emails, and the verification results are output. The multi-dimensional verification includes at least real email similarity verification and business boundary and consistency verification.

[0114] In this embodiment, at least two checks are performed on the generated initial training email to determine whether the initial training email meets the quality, security and compliance requirements.

[0115] Real email similarity verification involves comparing the initial training email with real emails in the organization's neighborhood to prevent excessive or near-abstract duplication of the real email text. Similarity calculation can employ the vector cosine similarity method, which involves converting the anonymized reference representations (e.g., anonymized text, text fingerprints, or vector indices) of both the initial training email and the real emails into vectors and calculating the cosine similarity between them. It's important to note that the real email similarity verification does not use the original text of the real emails, but rather their anonymized reference representations. Furthermore, the initial training email is only compared with these anonymized reference representations; the actual text is not provided to the large language model. If the similarity between the initial training email and any real email exceeds a preset threshold, it means that the email is too similar to a real email, posing a privacy risk and rendering the training meaningless. In this case, the verification result indicates that the similarity verification has failed.

[0116] Business boundary and consistency verification involves checking whether the business system, business action, process stage, subject type, displayed sender name, and displayed sender email in the initial training email are within the allowed range, based on the list of allowed fields, business scenario feature information, and the displayed sender identity in the target constraints, and determining whether they are consistent with the recent business behavior of the target personnel. For example, if the list of allowed fields in the target constraints does not include the business action "financial reimbursement," but the initial training email contains the content "Please complete the reimbursement approval as soon as possible," then the initial training email exceeds the business boundary, and the verification result indicates that the business boundary and consistency verification has failed. Similarly, if the initial training email uses a displayed sender name or displayed sender email that is inconsistent with the displayed sender identity specified in the target constraints, it is also determined that the business consistency verification has failed.

[0117] Finally, if the verification results indicate that the initial training email passes the multi-dimensional verification, the initial training email is identified as the target phishing training email.

[0118] In this embodiment, if the initial training email passes the real email similarity check and business boundary and consistency check (and other possible check items), the training entry placeholder in the initial training email is replaced with a unique link or identifier generated by the training platform to obtain the target phishing training email. The target phishing training email can be output in the form of a target email object, which may include at least: training task number, target personnel identifier, recipient email address, displayed sender name, displayed sender email address, email subject, email body, training entry point, structured context number, candidate email number, and check result.

[0119] Based on the above method, by determining business scenario feature information based on the target personnel's intranet access events, the subject of the training emails is no longer randomly selected manually or freely fabricated by the large model, but is derived from the intranet business systems that the target personnel have recently accessed, thus ensuring the relevance between the training emails and the target personnel's actual business activities. Simultaneously, by first identifying and de-identifying sensitive entities in the organizational neighborhood email corpus, and then extracting email corpus feature information, the subsequent email generation process will not access any sensitive data in the original email body, controlling the risk of privacy leakage at the source. By transforming business scenario feature information and email corpus feature information into structured target constraints, clear content boundaries are defined for the generation process of the large language model, avoiding the possibility of business system fabrication, business process errors, and the introduction of sensitive information that might result from the model's free interpretation. Multi-dimensional verification after generation, especially verification of similarity to real emails and verification of business boundaries and consistency, further ensures that the output emails do not copy real emails and do not exceed the business scope, achieving controllable email similarity and verifiable business consistency. Therefore, this embodiment can generate a phishing exercise email that is business-related, has controllable sensitive information risk, and verifiable similarity in an enterprise intranet environment, based on the actual business behavior of the target personnel and the email expression habits of their organization.

[0120] It should be noted that the specific data source types, feature extraction methods, constraint generation methods, and verification methods involved in the above embodiments are merely illustrative examples and do not constitute a limitation on the scope of protection of this application. For example, intranet access events can also come from other data sources that can reflect the business behavior of target personnel, such as application access logs, API gateway logs, terminal security logs, or work order flow logs.

[0121] In an optional embodiment, the specific implementation method of "determining the business scenario feature information of the initial exercise email based on the intranet access events of the target personnel and preset evaluation indicators" in the above embodiment will be described in detail.

[0122] See Figure 3This is a flowchart illustrating a method for determining business scenario feature information provided in an embodiment of this application. Figure 3 As shown, the method includes steps 301 to 303.

[0123] Step 301: The phishing exercise email generation device determines at least one business system accessed by the target personnel based on intranet access events.

[0124] Step 302: The phishing exercise email generation device determines the popularity value of each business system according to the preset evaluation indicators.

[0125] Step 303: The phishing exercise email generation device determines the business scenario feature information based on the business system with the highest popularity value.

[0126] In this embodiment, the preset evaluation indicators specifically include access frequency, recent access time, system sensitivity level, and operation type. Of course, in other embodiments, other dimensions can be added or replaced according to actual needs, such as access result type, operation time period distribution, etc.

[0127] First, identify at least one business system accessed by the target user based on intranet access events.

[0128] In this embodiment, the system aggregates intranet access events according to the target user's user ID and a preset statistical time window (e.g., the most recent 14 days, which can be configured), identifying which business systems the target user accessed within the statistical time window. Each intranet access event may contain a system identifier field (system_id) and a system name field (system_name). By deduplicating and aggregating based on these fields, the set of business systems accessed by the target user can be obtained.

[0129] Then, the popularity value of each business system is determined according to the preset evaluation indicators.

[0130] In this embodiment, for each business system, the system calculates an access frequency score, a recent access score, a system sensitivity level score, and an operation type score. Wherein:

[0131] The access frequency score can be expressed as the ratio of the number of times the business system is accessed within the statistical time window to the number of times the target user accesses the business system with the highest number of accesses within the same statistical time window.

[0132] For example, access frequency score ( The following formula (1) is used to calculate:

[0133] (Formula 1)

[0134] in, To determine when target personnel access the business system within the statistical time window T. Number of times, This refers to the number of times the target user accesses the business system most frequently within the statistical time window T.

[0135] The most recently visited score can be calculated based on the number of days since the target user last visited the business system, using either a linear decay or piecewise decay function, to make the most recently visited system score higher.

[0136] For example, the most recent access time score ( The following formula (2) is used to calculate:

[0137] (Formula 2)

[0138] in, For business systems The number of days since the last visit, where T is the length of the statistical time window.

[0139] The system sensitivity level score can be obtained by mapping the security sensitivity level of the business system in the enterprise's classification and filing. The higher the sensitivity level, the higher the score.

[0140] For example, the system sensitivity level score ( The system is mapped according to the enterprise's classification and filing configuration: low-sensitivity systems (Level 1 systems) are mapped to 0.30, medium-sensitivity systems (Level 2 systems) are mapped to 0.60, high-sensitivity systems (Level 3 systems) are mapped to 0.85, and core-sensitivity systems (Level 4 systems) are mapped to 1.00.

[0141] The operation type score can be determined according to the preset operation type weight mapping table. When there are multiple operation types on the business system within the statistical time window, the highest score among the operation types in the system can be taken.

[0142] For example, the operation type score ( Based on a preset operation weight mapping table, when multiple operation types exist within a statistical time window, the highest operation score of the system can be used. An example operation weight configuration is as follows: Login, normal browsing, and list viewing correspond to 0.20; details viewing, status query, and notification viewing correspond to 0.40; material supplementation, form editing, and draft saving correspond to 0.60; process submission, approval confirmation, permission renewal, and permission change correspond to 0.85; batch download, data export, release and deployment, and configuration change correspond to 1.00.

[0143] It is understood that the specific calculation methods of the above four scores are only examples. Other methods can also be used in other embodiments. For example, the access frequency score can also be the proportion of the number of times the business system is accessed to the total number of times it is accessed, and the recent access score can also be exponentially decayed.

[0144] After obtaining the scores from the four dimensions mentioned above, the scores can be weighted and summed according to preset weights to obtain the popularity value of each business system. In practical applications, the weights can be configured according to specific implementation needs. For example, the access frequency weight can be 0.30, the recent access weight can be 0.25, the system sensitivity level weight can be 0.25, and the operation type weight can be 0.20. However, the weight configuration is not limited to these and can be adjusted according to the actual exercise strategy.

[0145] Finally, business scenario characteristic information is determined based on the business system with the highest popularity value.

[0146] In this embodiment, the business system with the highest popularity value is selected as the priority business scenario source. For example, if the popularity value of "Permission Management Platform" is 0.85, which is higher than that of "Code Repository Platform" (0.60) and "Work Order System" (0.55), then "Permission Management Platform" is selected as the priority business scenario. Further, the internal network access event input intent recognition component (which can be implemented based on a rule mapping table or by calling a large language model for auxiliary classification) from the business system with the highest popularity value, after being cleaned and structured, determines the process stage of the target user within that business system based on the corresponding operation type and access path, and generates email subject suggestions and business action candidate sets matching that process stage, thereby obtaining business scenario feature information. In a specific example, if the target person's operation type on the "Permission Management Platform" includes "Permission Renewal Submission" and "Request Permission to View", the system determines that the person may be in the process stage of "Permission Renewal Submission After / Approval Confirmation" based on the mapping relationship between operation type and process stage, and generates business action candidates such as "Permission Review Reminder" and "Supplementary Material Submission Reminder" and corresponding email subject suggestions accordingly.

[0147] This embodiment uses four preset evaluation indicators—access frequency, recent access time, system sensitivity level, and operation type—to calculate the popularity score of each business system. The system with the highest popularity score is selected as the source of the business scenario. This ensures that the determination of the business scenario no longer relies on manual experience or random selection, but is automatically derived based on the target personnel's recent real-world business access behavior data. This approach ensures that the business themes surrounding the exercise emails are highly relevant to the target personnel's actual work content, while prioritizing business scenarios with higher security sensitivity, thereby improving the effectiveness and relevance of the exercise.

[0148] In one optional embodiment, the specific implementation of "generating email corpus feature information based on email corpus of the target person's organizational neighborhood" in the above embodiment will be described in detail.

[0149] In this embodiment, sensitive entity identification and de-identification processing can be performed on the email corpus, and email corpus feature information can be extracted. Specifically, the email corpus feature information includes the distribution of topic structure, body structure, tone style, business actions, and body length. This distribution information describes the statistical patterns of emails within the organization's domain at the expression level. For example, when writing notification emails, personnel within the organization tend to use a topic structure of "greeting + description + deadline + contact person," a formal but not abrupt tone, and a body length concentrated between 150 and 350 characters, etc.

[0150] See Figure 4 This is a flowchart illustrating a method for generating email corpus feature information according to an embodiment of this application. Figure 4 As shown, the method includes steps 401 to 403.

[0151] Step 401: The phishing exercise email generation device performs sensitive entity recognition on the email corpus.

[0152] Step 402: The phishing exercise email generation device performs anonymization processing on the email corpus.

[0153] Step 403: The phishing exercise email generation device extracts features from the desensitized email text or structured intermediate results to generate email corpus feature information.

[0154] First, sensitive entity recognition is performed on the email corpus.

[0155] In this embodiment, sensitive entity recognition refers to the automatic detection and labeling of sensitive information in email corpora. The types of sensitive entities may include, but are not limited to: real personnel names, customer names, project codes, contract numbers, order numbers, monetary amounts, mobile phone numbers, ID card numbers, email accounts, internal network IP addresses, real domain names, real attachment names, access credentials, tokens, password-like fields, and confidential document titles. Sensitive entity recognition can be implemented in various ways, such as using one or more combinations of regular expression matching, dictionary matching, named entity recognition models, or large language models for extraction. In one implementation, the system can prioritize rule matching (regular expressions and dictionaries) for rapid identification, and then use named entity recognition models or large language models for supplementary identification of entity types that cannot be covered by the rules.

[0156] Secondly, after completing the identification of sensitive entities, the email corpus is anonymized.

[0157] In this embodiment, desensitization refers to processes such as replacing identified sensitive entities with corresponding entity category labels (e.g., replacing "Zhang San" with "<person's name>", and "HT-2026-00123" with "<contract number>"), deletion, generalization, or hashing, so that the desensitized text no longer contains any of the original real sensitive values. The desensitized text retains only abstract information such as the email's structure, tone, and expression, while the original sensitive information has been completely stripped away. It is understood that the specific desensitization method can be flexibly chosen according to the security policy, as long as it ensures that the desensitized text does not leak the original sensitive information.

[0158] Next, feature extraction is performed on the de-identified email text or structured intermediate results to generate email corpus feature information.

[0159] In this embodiment, the specific process of feature extraction may include: extracting the topic structure (such as the paragraph organization and logical hierarchy of the email), body structure (such as the arrangement of the opening, body and closing of the email), tone style (such as formal / informal, imperative / consultative, etc.), business actions (such as notification, reminder, confirmation, request for reply, etc.), and body length (such as the number of characters or words in the email body) from each de-identified email within the organization's neighborhood. Then, the extraction results of all de-identified emails are statistically summarized to obtain the distribution of topic structure, body structure, tone style, business actions, and body length. Optionally, a large language model can be called to assist in text structure analysis, tone classification and business action annotation, but the large language model only receives text that has already been de-identified or structured intermediate results and will not access the original email body.

[0160] This embodiment achieves two beneficial effects by first identifying and de-identifying sensitive entities in the email corpus of the organization's neighborhood, and then extracting abstract feature information. On the one hand, sensitive information such as customer names, contract numbers, and amounts in the original email body is completely removed during the de-identification stage, and all subsequent feature extraction and email generation processes will not come into contact with any real sensitive data, thus ensuring privacy and security at the data flow level. On the other hand, the extracted abstract features such as the distribution of topic structure, body structure, and tone style can accurately reflect the email expression habits of the organization's neighborhood, providing a reliable basis for generating writing constraints that fit the actual communication style of the target personnel's organization.

[0161] In an optional embodiment, this embodiment further illustrates that the target constraint may also include revealing the sender's identity, and describes in detail the process of determining the sender's identity.

[0162] See Figure 5 This is a flowchart illustrating a method for determining the sender's identity, provided in an embodiment of this application. Figure 5 As shown, the method includes steps 501 to 503.

[0163] Step 501: The phishing exercise email generation device obtains the target person's email communication metadata, which includes at least one of the following: sender, recipient, sending time, sender identity type, email subject classification, and associated business system.

[0164] Step 502: The phishing exercise email generation device constructs a communication relationship diagram centered on the target person based on the email communication metadata. The communication relationship diagram includes the communication relationship between the target person and at least one candidate sender identity.

[0165] Step 503: The phishing exercise email generation device determines the comprehensive matching degree of each candidate sender identity based on the communication relationship diagram, and determines the sender identity to be displayed from each candidate sender identity based on the comprehensive matching degree.

[0166] In phishing drill emails, the displayed sender identity (i.e., the "sender" name and email address seen by the recipient) is a crucial factor affecting the email's credibility. If the displayed sender identity does not match the business scenario of the email—for example, if the email content is a reminder about permission renewal, but the displayed sender identity is a "financial reimbursement notification"—the recipient can easily identify the email as suspicious, significantly diminishing the drill's effectiveness. Therefore, this embodiment provides a technical solution based on a communication relationship graph to automatically determine the displayed sender identity.

[0167] First, obtain the target person's email communication metadata.

[0168] In this embodiment, email communication metadata refers to structured data describing the recent email communication behavior of the target personnel. This data may include at least one of the following: sender, recipient, sending time, sender identity type, email subject category, and associated business systems. The system can preferentially obtain this communication metadata from email gateway logs or the email auditing system, or it can obtain it from the send and receive records of the enterprise email system.

[0169] Secondly, based on email communication metadata, a communication relationship diagram centered on the target personnel is constructed.

[0170] In this embodiment, the communication relationship graph is a graph data structure used to represent the recent communication relationship network between the target personnel and various email sending identities. In the communication relationship graph, nodes can include target personnel nodes, system notification email nodes (such as permission management platform notification emails, code repository platform notification emails), departmental public email nodes, project team email nodes, and general exercise email nodes, etc.; the edges connecting each node represent email communications that have occurred between the target personnel and the corresponding email identities within a statistical time window. Each edge can record the following information: number of communications, most recent communication time, email subject category, associated business system, and identity type. Through the communication relationship graph, the system can intuitively view which email identities the target personnel have recently communicated with, as well as the frequency and business association of these communications.

[0171] Next, the overall matching degree of each candidate sender identity is determined based on the communication relationship diagram, and the sender identity to be displayed is determined from each candidate sender identity based on the overall matching degree.

[0172] In this embodiment, all email identities that have communication relationships with the target person can be extracted from the communication relationship graph as candidate sender identities, and then a comprehensive matching degree can be calculated for each candidate sender identity.

[0173] The overall matching score is calculated by weighting three dimensions: communication frequency score, recent communication score, and business scenario matching score. Specifically, the communication frequency score measures how frequently the target person communicates with a candidate identity or a similar identity within the statistical time window; the more frequent the communication, the higher the score. The recent communication score measures whether the candidate identity has recently communicated with the target person; the more recent the communication, the higher the score. The business scenario matching score measures whether the candidate's displayed sending identity is suitable for the current business scenario. The business scenario matching score can be calculated with reference to the preset "Business System - Display Sender Identity" mapping table. For example, if the current business scenario is "Permission Management Platform" and the candidate identity happens to be "Permission Management Platform Notification Email", then the business scenario matching score can be the highest level (e.g., 1.00); if the candidate identity is strongly related to the current business process but not directly corresponding, the next highest level can be taken (e.g., 0.80); if the candidate identity is weakly or indirectly related to the current business process, the middle level can be taken (e.g., 0.70); if the candidate identity is only a notification identity from the same project group or department, the lower level can be taken (e.g., 0.60); if the candidate identity is a general exercise notification identity, the lower level can be taken (e.g., 0.40); if the candidate identity is completely unrelated to the current business, the score can be 0. Optionally, for candidate identities that cannot be directly matched through the preset mapping table, the de-identified topic categories, identity types, and business scenarios can be used for auxiliary classification, but the classification results still need to be mapped to the above preset score levels.

[0174] After obtaining the communication frequency score, recent communication score, and business scenario matching score, the three are weighted and summed according to preset weights to obtain the comprehensive matching degree of each candidate sender identity. Then, the candidate sender identity with the highest comprehensive matching degree is selected as the displayed sender identity, and this displayed sender identity is written into the target constraints. The displayed sender identity can include the display name (e.g., "Permission Management Platform Notification"), the display email (e.g., "iam-notify@company.com"), the identity category (e.g., "System Notification"), and the matching business scenario information.

[0175] It should be noted that for target personnel with insufficient communication records (such as newly hired employees), the communication relationship diagram may lack enough candidate sender identities. In this case, the system can downgrade the process, for example, by directly using the default notification identity of the target business system, the public email identity of the target personnel's department, or the general security exercise identity as the displayed sender identity.

[0176] This embodiment constructs a communication relationship graph based on email communication metadata and calculates a comprehensive matching degree based on three dimensions: communication frequency, recent communication time, and business scenario matching degree to select the sender identity to be displayed. This ensures that the final selected sender identity not only matches the target person's recent communication habits but also remains consistent with the current business scenario, thereby effectively improving the credibility of the exercise emails without exposing the real communication content.

[0177] In an optional embodiment, the multidimensional verification further includes writing structure verification and sensitive information verification. In addition to real email similarity verification and business boundary and consistency verification, the multidimensional verification process further includes writing structure verification and sensitive information verification. The above embodiments have already described real email similarity verification and business boundary and consistency verification; the following will describe writing structure verification and sensitive information verification one by one:

[0178] Writing structure verification involves the system checking whether the initial practice email conforms to the expression patterns abstracted from the anonymized corpus of the organizational domain, based on writing constraints. Specifically, it checks whether the subject structure of the initial practice email is consistent with the subject structure pattern specified in the writing constraints, whether the body structure follows the preset paragraph organization, whether the tone and style conform to the statistical preferences of the organizational domain (e.g., whether the formality and politeness of the email are within the allowable range of the writing constraints using a tone classification model), and whether the body length is within the preset length range. By verifying the writing structure, the system can avoid generating emails that significantly deviate from the communication habits of the organizational domain in terms of expression, thereby reducing the risk of the recipient recognizing the error at a glance.

[0179] Sensitive information verification includes the system checking whether the initial training email contains sensitive content types specified in the prohibited content rule list. The prohibited content rule list, generated in the above embodiments, defines content types that must not appear in emails, such as customer name entities, contract number entities, application form number entities, account number entities, password / credential entities, actual attachment name entities, and amount entities. Sensitive information verification can be performed using regular expression matching, keyword matching, entity recognition models, etc., with the aim of ensuring that the large language model does not fabricate any sensitive formats or introduce real sensitive information during the generation process.

[0180] See Figure 6 This is a flowchart illustrating a multi-dimensional verification method provided in an embodiment of this application. Figure 6 As shown, the method includes steps 601 to 603.

[0181] Step 601: The phishing exercise email generation device performs real email similarity verification, business boundary and consistency verification, writing structure verification and sensitive information verification on the initial exercise email, generating multiple single-dimensional verification results.

[0182] Step 602: If multiple single-dimensional verification results indicate that the phishing exercise email generation device has passed the multi-dimensional verification, the phishing exercise email generation device determines that the initial exercise email has passed the multi-dimensional verification.

[0183] Step 603: If any single-dimensional verification result fails, the phishing exercise email generation device determines that the initial exercise email has failed the multi-dimensional verification.

[0184] First, the initial training emails are subjected to the following checks: similarity to real emails, business boundary and consistency, writing structure, and sensitive information. Multiple single-dimensional check results are generated. Each single-dimensional check result can independently represent whether the check for that dimension "passed" or "failed." Optionally, the above four checks can be executed simultaneously in parallel according to a preset configuration to improve check efficiency. In another optional embodiment, the checks can also be executed sequentially according to a preset order. For example, sensitive information check and business boundary and consistency check can be executed first. If either of these two fails, the system can directly reject the candidate email without waiting for other checks to complete, thereby saving computational resources.

[0185] Secondly, if multiple single-dimensional verification results indicate that the initial training email has passed the multi-dimensional verification, then the initial training email is determined to have passed the multi-dimensional verification. That is, the initial training email is only considered to have passed the multi-dimensional verification if all four verifications (real email similarity verification, business boundary and consistency verification, writing structure verification, and sensitive information verification) return a "pass" result.

[0186] If any single-dimensional verification result fails, the initial training email is determined to have failed the multi-dimensional verification. For example, if the initial training email is detected to contain text in the format of a contract number during the sensitive information verification (even if the contract number is not a real contract number), the sensitive information verification result is "failed," and the entire multi-dimensional verification result is "failed."

[0187] In addition, in an optional embodiment, a handling mechanism is provided after verification failure. If the verification result indicates that the initial training email has failed the multidimensional verification, the initial training email is marked as verification failed, and candidate training emails are regenerated based on the target constraints.

[0188] Specifically, initial training emails that fail to pass verification can be marked as failing, and detailed reasons for failure can be recorded (such as which prohibited content rule was hit, which real email had a similarity exceeding the threshold, which allowed field was exceeded, which writing constraint was violated, etc.). Then, the failure reason information can be fed back to the email generation step, and new candidate training emails can be generated again by calling the large language model based on the same target constraints.

[0189] Optionally, during regeneration, failure reason information can be appended to the target constraints as an additional negative example to reduce the probability of similar failures recurring. In another optional implementation, if all checks still fail after a preset number of regenerations, the system can enter a manual review process, where security operations personnel manually review and adjust the email content.

[0190] This embodiment expands the scope of multi-dimensional verification from similarity and business boundaries to writing style and sensitive information control by introducing writing structure verification and sensitive information verification, forming a more comprehensive quality assurance system. Simultaneously, through a clear "all passes are required" decision logic and an automatic regeneration mechanism after verification failure, the generation process of phishing drill emails possesses closed-loop self-control capabilities. It can automatically filter out unqualified candidate emails and continuously improve output quality through regeneration, thereby effectively reducing manual review costs.

[0191] See Figure 7 The diagram illustrates the overall architecture of a phishing drill email generation system provided in an embodiment of this application. Figure 7 As shown, the system includes seven functional modules, and the inputs, processing logic and outputs of each module are described below.

[0192] Internal network access event processing module: Taking unified authentication logs, VPN logs, zero-trust gateway audit records, bastion host operation logs, business system audit logs, and business system asset ledgers or URL path mapping tables as inputs, it pulls internal network access events according to target personnel and statistical time windows, cleans, deduplicates, and aggregates the access events, calculates the popularity value of each business system according to the preset evaluation indicators, determines the process stage of the target personnel, generates a set of business action candidates, and outputs business scenario feature information.

[0193] The communication relationship processing module takes email gateway communication metadata, email audit metadata, organizational structure information, and a pre-configured list of sender identities as input. Based on the email communication metadata of the target personnel, it constructs a communication relationship graph. For each candidate sender identity in the communication relationship graph, it calculates the communication frequency score, the most recent communication score, and the business scenario matching score. The scores are then weighted and summed according to preset weights to obtain the comprehensive matching degree. Based on the comprehensive matching degree, the sender identity to be displayed is determined from each candidate sender identity.

[0194] Organizational Neighborhood Corpus Processing Module: Taking organizational structure, project teams, roles, email group information, and organizational neighborhood email corpus as input, it determines the organizational neighborhood of the target personnel, obtains email corpus samples that meet the conditions, performs sensitive entity identification and de-identification processing on the email corpus, clusters and statistically analyzes the de-identified text, and extracts email corpus feature information such as topic structure distribution, body structure distribution, tone style distribution, business action distribution, and body length distribution.

[0195] The structured controlled business context generation module takes business scenario feature information, sender identity, email corpus feature information, and strategy configuration as input, integrates the outputs of the aforementioned modules into target constraints, including a list of allowed fields, a list of prohibited content rules, writing constraints, and candidate email verification rules; it prohibits the original email body and truly sensitive fields from entering the large language model input, and outputs the structured controlled business context (i.e., the target constraints).

[0196] Candidate email generation module: Taking the structured controlled business context and the large language model configuration parameters as input, it calls the large language model only based on the target constraints and generates one or more initial exercise emails according to the list of allowed fields, the list of prohibited content rules and writing constraints; the initial exercise emails only contain exercise entry placeholders and do not directly generate real exercise links.

[0197] Candidate email verification module: Taking the initial training email, target constraints, sensitive entity identification rules, similarity model, and organizational neighborhood reference samples as input, it performs sensitive information verification, real email similarity verification, business boundary and consistency verification, and writing structure verification on the initial training email; it outputs the verification results, the output initial training email, or the reason for failure.

[0198] The target email output and auditing module takes the verified initial exercise email, exercise task configuration, target personnel information, and verification results as input. If the verification results indicate that the initial exercise email has passed multi-dimensional verification, the exercise entry placeholder is replaced with a unique link or identifier bound to the exercise task and target personnel. The initial exercise email is identified as the target phishing exercise email and encapsulated for output. At the same time, an audit record containing a structured context number, various verification results, the highest similarity to the real email, and the output time is generated.

[0199] See Figure 8 The diagram illustrates the overall process of a phishing drill email generation method provided in an embodiment of this application. Figure 8 As shown, the method includes eight steps, from identifying target personnel to outputting target phishing exercise emails and audit logs, forming a complete processing loop.

[0200] Step 1: Identify the target personnel for the exercise.

[0201] This embodiment supports two methods: First, import the target personnel's email list through the front-end page, obtain department, position, and project team information from the user center, and compare it with historical exercise records to obtain historical exercise results and risk levels; second, divide all personnel into four quadrants according to the scope of risk impact and the probability of risk occurrence, and automatically sample from each quadrant according to a preset sampling ratio to obtain the target personnel list. Taking target personnel U1 as an example, their user identifier is U1, their user email is U1@company.com, their department is the platform R&D department, their position is backend R&D engineer, their project team is the permission management platform project team, their historical exercise result is no participation in the last 90 days, and their risk level is medium risk.

[0202] Step 2: Obtain intranet access events and determine business scenario characteristics.

[0203] In this embodiment, the system queries intranet access records based on the target personnel's email address and a preset statistical time window. It prioritizes rule matching based on the system asset ledger and URL path mapping table to identify the business system, system type, and system sensitivity level corresponding to each access record. The intranet access event list is aggregated by business system dimension. The system calculates the popularity value of each business system according to the preset evaluation indicators, selects the business system with the highest popularity value, and performs intent recognition by combining operation type and access path to determine the process stage of the target personnel and output business scenario feature information.

[0204] Step 3: Construct a communication relationship diagram and determine the sender's identity.

[0205] In this embodiment, the system obtains recent email communication metadata (including fields such as sender, recipient, sending time, sender identity type, email subject classification, and associated business system) of the target person from the email gateway or email auditing system, and constructs a communication relationship diagram with the target person as the central node; calculates the comprehensive matching degree for each candidate sender identity in the communication relationship diagram, and determines the sender identity to be displayed from each candidate sender identity based on the comprehensive matching degree.

[0206] Step 4: Determine the organizational neighborhood and extract feature information from the email corpus.

[0207] In this embodiment, the system determines the organizational neighborhood based on the target personnel's department, project team, position, and email group, and obtains email corpus samples according to the statistical time period and sample upper limit; it performs sensitive entity identification and de-identification processing on the email corpus, and extracts email corpus feature information such as topic structure distribution, body structure distribution, tone style distribution, business action distribution, and body length distribution from the de-identified text; the original email body does not enter the subsequent steps.

[0208] Step 5: Generate a structured controlled business context (i.e., target constraints) based on business scenario feature information and email corpus feature information.

[0209] In this embodiment, the system integrates business scenario feature information, display sender identity, and email corpus feature information into target constraints, including a list of allowed fields, a list of display sender identity, a list of prohibited content rules, writing constraints, and candidate email verification rules.

[0210] Step 6: Use the large model to generate candidate emails (i.e., initial practice emails).

[0211] In this embodiment, the system uses the target constraints as input to the large language model to generate an initial training email. The initial training email is encapsulated as a candidate email object, which includes at least the candidate email number, subject, display sender name, display sender email address, body, training entry placeholder, business action used, and structured context number.

[0212] Step 7: Perform candidate email verification (i.e., multi-dimensional verification).

[0213] In this embodiment, the system performs sensitive information verification, business boundary and consistency verification, writing structure verification, and real email similarity verification on the initial training email, and outputs multiple single-dimensional verification results.

[0214] Step 8: Output the target email (i.e., the target phishing exercise email) and audit log.

[0215] In this embodiment, if the verification result indicates that the initial training email passes the multi-dimensional verification, the training entry placeholder is replaced with a unique link or identifier generated by the training platform, the initial training email is identified as the target phishing training email and output, and an audit record is generated at the same time; if the verification result indicates that the initial training email fails the multi-dimensional verification, the reason for failure is recorded, and a candidate training email is regenerated based on the target constraint conditions.

[0216] See Figure 9 This illustration shows a flowchart of a process for determining business scenario feature information based on the intranet access events of target personnel and preset evaluation indicators, according to an embodiment of this application. Figure 9 As shown, the process includes four stages: aggregation, scoring, weighted summation, and selection.

[0217] Aggregation Phase: Within a preset statistical time window, the system aggregates intranet access events by business system dimension, obtaining the access count, most recent access time, system sensitivity level, and operation type set for each business system. Taking target user U1 as an example, the aggregation results are: Access Management Platform – 2 accesses, most recent access time D-1 (core sensitive); Work Order System – 1 access, most recent access time D-2 (medium sensitive); Code Repository Platform – 1 access, most recent access time D-3 (medium sensitive); Build and Release Platform – 1 access, most recent access time D-4 (core sensitive).

[0218] Scoring phase: The system calculates the sub-scores of each business system according to the four dimensions of the preset evaluation indicators.

[0219] Weighted Summation Phase: The system performs a weighted summation of the scores across the four dimensions according to preset weights. In the example, the calculated popularity values ​​for each business system are: Permission Management Platform H=0.964, Build and Release Platform H=0.683, Work Order System H=0.628, and Code Repository Platform H=0.607.

[0220] Selection and Intent Recognition Phase: The system determines business scenario feature information based on the business system with the highest popularity value. In the example, the system selects the permission management platform with the highest popularity value (H=0.964), and performs intent recognition by combining the operation type and access path mapping table. It determines that the target person U1 is in the "permission renewal submission / approval confirmation" process stage, and generates a set of business action candidates (permission review, permission renewal confirmation, application material supplement reminder). The final output business scenario feature information includes: business system type is permission management system, business action is permission renewal and permission review, and business process stage is permission renewal submission / approval confirmation.

[0221] See Figure 10This illustration shows a flowchart provided by an embodiment of this application, illustrating a process for constructing a communication relationship graph based on email communication metadata, determining the comprehensive matching degree of each candidate sender's identity, and determining the sender's identity to be displayed. Figure 10 As shown, the process includes four stages: extraction of email communication metadata, construction of communication relationship graph, calculation of comprehensive matching degree, and display of sender identity selection.

[0222] Email communication metadata extraction stage: The system obtains the target personnel's email communication metadata from the email gateway or email auditing system. Fields include sender (sender, used to identify which system emails, department emails or project team emails the target personnel have recently contacted), recipient (recipient), send_time (sending time, used to calculate the recent communication score), mail_direction (email direction), sender_type (sender identity type), subject_label (email subject classification, used to determine whether the communication content is related to the current business scenario), related_system (related business system, used to determine whether the email identity matches the current business system), and sender_domain (sender domain), etc.

[0223] Communication Relationship Diagram Construction Phase: The system constructs a communication relationship diagram with the target personnel as the central node and various email identities (system notification email, departmental public email, project team email, or general exercise email) as surrounding nodes. The diagram records information such as the number of communications, the most recent communication time, the topic category, associated business systems, and the identity type. In the example, the communication relationship diagram includes: U1 and the permission management platform notification email: 8 communications, most recent communication 1 day ago, topic: permission application / permission renewal; U1 and the work order system notification email: 5 communications, most recent communication 3 days ago; U1 and the project team public email: 4 communications, most recent communication 7 days ago; U1 and the security exercise notification email: 1 communication, most recent communication 20 days ago.

[0224] In the comprehensive matching degree calculation stage, the system calculates the communication frequency score, the most recent communication score, and the business scenario matching score for each candidate sender identity in the communication relationship graph, and sums them up according to preset weights to obtain the comprehensive matching degree.

[0225] In the sender identity selection phase, the system selects the candidate sender identity with the highest overall matching score as the displayed sender identity. In the example, the notification email address of the permission management platform (display name "Permission Management Platform Notification", display email address "iam-notify@company.com", overall matching score 0.96) is selected as the displayed sender identity, and target constraints are written to constrain the displayed sender name and displayed sender email address in the initial exercise email.

[0226] See Figure 11 This illustration shows a flowchart of a process for sensitive entity identification and de-identification of email corpora from the organizational domain of a target individual, as provided in an embodiment of this application, and for extracting feature information from the email corpora. Figure 11 As shown, the process includes six stages: determining the organizational neighborhood, acquiring the corpus, identifying sensitive entities, desensitizing the data, extracting features, and summarizing statistics.

[0227] Organizational neighborhood determination and corpus acquisition stage: The system determines the scope of the organizational neighborhood based on the target personnel's department, project team, position, email group, and business system user group. Email corpus samples are acquired according to the statistical time period (e.g., the most recent 60 days), organizational neighborhood scope, email type, and sample limit. The original emails are only used for subsequent sensitive entity identification and de-identification processing within the controlled security domain.

[0228] Sensitive entity identification and desensitization stage: The system performs sensitive entity identification on the email corpus. The identification objects include real personnel names, customer names, project codes, contract numbers, order numbers, amounts, mobile phone numbers, ID card numbers, email accounts, internal network IP addresses, real domain names, real attachment names, and password fields, etc. After the identification is completed, the sensitive entities are desensitized by replacing the real sensitive values ​​with entity category tags or deleting them, so that the desensitized text does not contain any real sensitive values.

[0229] Feature Extraction and Statistical Summary Stage: The system extracts feature information from the anonymized text of emails, including the distribution of subject structure, body structure, tone style, business actions, and body length. In the example, the subject of an original email is "[Permission Management Platform] Please supplement the application materials for production environment permissions for XX customer." After anonymization, the extracted features are: subject structure "system name + business action + reminder," body structure "salutation + background description + pending matter + time reminder + system entry reminder," tone style "formal, concise, slightly urgent," and business actions "supplement materials, resubmit for approval, check status." The system summarizes and statistically analyzes the features of all emails within the organization's domain to obtain email corpus feature information. For example: subject structure A (system name + business action + reminder) accounts for 42%, body structure A (matter background + pending action + time reminder + system entry) accounts for 51%, tone style "formal, concise, neutral" accounts for 63%, and body length 50-120 characters accounts for 46%. The original email body is not included in subsequent steps.

[0230] As can be seen, the above mainly describes the solutions provided by the embodiments of this application from a methodological perspective. To achieve the above functions, the embodiments of this application provide corresponding hardware structures and / or software modules for executing each function. Those skilled in the art should readily recognize that, in conjunction with the modules and algorithm steps of the various examples described in the embodiments disclosed herein, the embodiments of this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0231] This application embodiment can divide the phishing exercise email generation device into functional modules according to the above method example. For example, each function can be divided into a separate functional module, or two or more functions can be integrated into one processing module. The integrated module can be implemented in hardware or as a software functional module. Optionally, the module division in this application embodiment is illustrative and only represents one logical functional division; other division methods may be used in actual implementation.

[0232] In some embodiments, this application also provides a phishing practice email generation apparatus. This phishing practice email generation apparatus may include one or more functional modules for implementing the phishing practice email generation method of the above method embodiments.

[0233] For example, Figure 12 This is a schematic diagram of a phishing practice email generation device provided in an embodiment of this application. Figure 12 As shown, the phishing exercise email generation device 1200 includes: a determination module 1201, a generation module 1202, and a verification module 1203.

[0234] The determination module 1201 is used to determine the business scenario feature information of the initial exercise email based on the intranet access events of the target personnel and preset evaluation indicators, and to generate email corpus feature information based on the email corpus of the target personnel's organizational neighborhood.

[0235] The generation module 1202 is used to generate target constraints based on business scenario feature information and email corpus feature information. The target constraints include at least a list of allowed fields, a list of prohibited content rules, and writing constraints.

[0236] The generation module 1202 is also used to generate initial exercise emails based on target constraints;

[0237] The verification module 1203 is used to perform multi-dimensional verification on the initial exercise emails and output the verification results. The multi-dimensional verification includes at least real email similarity verification and business boundary and consistency verification.

[0238] The determination module 1201 is also used to determine the initial training email as the target phishing training email when the verification result indicates that the initial training email has passed the multi-dimensional verification.

[0239] In one optional embodiment, the preset evaluation indicators include access frequency, recent access time, system sensitivity level, and operation type;

[0240] Module 1201 is specifically used for:

[0241] Identify at least one business system accessed by the target user based on intranet access events;

[0242] The popularity value of each business system is determined based on preset evaluation indicators;

[0243] Determine the business scenario characteristics based on the business system with the highest popularity value.

[0244] In one optional embodiment, the email corpus feature information includes topic structure distribution, body structure distribution, tone style distribution, business action distribution, and body length distribution;

[0245] Module 1201 is specifically used for:

[0246] Sensitive entity identification and de-identification are performed on the email corpus, and feature information of the email corpus is extracted.

[0247] In an optional embodiment, the target constraint further includes revealing the sender's identity;

[0248] Module 1202 is also used for:

[0249] Obtain email communication metadata of the target personnel. Email communication metadata includes at least one of the following: sender, recipient, sending time, sender identity type, email subject category, and associated business system.

[0250] Based on email communication metadata, construct a communication relationship diagram centered on the target person. The communication relationship diagram includes the communication relationship between the target person and at least one candidate sender identity.

[0251] The overall matching degree of each candidate sender identity is determined based on the communication relationship diagram, and the sender identity to be displayed is determined from each candidate sender identity based on the overall matching degree.

[0252] In an optional embodiment, the multidimensional verification also includes writing structure verification and sensitive information verification;

[0253] Verification module 1203 is specifically used for:

[0254] The initial exercise emails were subjected to real email similarity verification, business boundary and consistency verification, writing structure verification and sensitive information verification, generating multiple single-dimensional verification results.

[0255] If multiple single-dimensional validation results all indicate that the initial training email has passed multi-dimensional validation; or

[0256] If any single-dimensional verification result fails, the initial training email is determined to have failed the multi-dimensional verification.

[0257] In an optional embodiment, the verification module 1203 is further configured to:

[0258] If the verification results indicate that the initial training email fails the multidimensional verification, the initial training email is marked as verification failed, and candidate training emails are regenerated based on the target constraints.

[0259] It should be noted that the phishing exercise email generation device can implement all the processes implemented in the above method embodiments and achieve the same beneficial effects. To avoid repetition, it will not be described again here.

[0260] In the case where the functions of the integrated modules described above are implemented in hardware, this application provides a possible structural schematic diagram of the electronic device involved in the above embodiments. For example... Figure 13 As shown, the electronic device 1300 includes: a processor 1302, a communication interface 1303, and a bus 1304. Optionally, the electronic device 1300 may also include a memory 1301.

[0261] Processor 1302 may implement or execute various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 1302 may be a central processing unit, a general-purpose processor, a digital signal processor, an application-specific integrated circuit, a field-programmable gate array, or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It may implement or execute various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 1302 may also be a combination that implements computing functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0262] The communication interface 1303 is used to connect to other devices via a communication network. This communication network can be Ethernet, wireless access network, wireless local area network (WLAN), etc.

[0263] The memory 1301 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto.

[0264] As one possible implementation, the memory 1301 can exist independently of the processor 1302. The memory 1301 can be connected to the processor 1302 via a bus 1304 and is used to store instructions or program code. When the processor 1302 calls and executes the instructions or program code stored in the memory 1301, it can implement the phishing exercise email generation method provided in this application embodiment.

[0265] In another possible implementation, the memory 1301 can also be integrated with the processor 1302.

[0266] The 1304 bus can be an Extended Industry Standard Architecture (EISA) bus, etc. The 1304 bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 13 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0267] Through the above description of the implementation methods, those skilled in the art can clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the phishing exercise email generation device can be divided into different functional modules to complete all or part of the functions described above.

[0268] This application embodiment also provides a chip, which includes a processor and a communication interface. The communication interface is coupled to the processor. The processor is used to run programs or instructions to implement the various processes of the above-described phishing exercise email generation method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0269] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

[0270] This application embodiment also provides a readable storage medium storing a program or instructions that, when executed by a computer, implement the phishing practice email generation method provided in the above embodiments. It is understood that all or part of the processes in the above method embodiments can be executed by computer instructions instructing related hardware; the readable storage medium can be any of the foregoing embodiments or memory; the readable storage medium can also be an external storage device of the phishing practice email generation device, such as a plug-in hard drive, Smart Media Card (SMC), Secure Digital (SD) card, flash card, etc., equipped on the phishing practice email generation device. Further, the readable storage medium can include both internal storage units of the phishing practice email generation device and external storage devices. The readable storage medium is used to store the computer program and other programs and data required by the phishing practice email generation device. The readable storage medium can also be used to temporarily store data that has been output or will be output.

[0271] This application also provides a computer program product, which is stored in a storage medium and, when executed by a computer, implements the phishing exercise email generation method provided in the above embodiments.

[0272] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0273] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a computer software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0274] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A method for generating phishing drill emails, characterized in that, include: The business scenario feature information of the initial exercise email is determined based on the intranet access events of the target personnel and the preset evaluation indicators, and the email corpus feature information is generated based on the email corpus of the target personnel's organizational neighborhood. Based on the business scenario feature information and the email corpus feature information, target constraints are generated. The target constraints include at least a list of allowed fields, a list of prohibited content rules, and writing constraints. Generate initial exercise emails based on the aforementioned target constraints; Perform multi-dimensional verification on the initial exercise emails and output the verification results. The multi-dimensional verification includes at least real email similarity verification and business boundary and consistency verification. If the verification result indicates that the initial training email passes the multi-dimensional verification, the initial training email is identified as a target phishing training email.

2. The method for generating phishing practice emails according to claim 1, characterized in that, The preset evaluation indicators include access frequency, recent access time, system sensitivity level, and operation type; The business scenario characteristic information for determining the initial exercise email based on the target personnel's intranet access events and preset evaluation indicators includes: Based on the intranet access events, determine at least one business system accessed by the target personnel; The popularity value of each business system is determined according to the preset evaluation indicators; The business scenario feature information is determined based on the business system with the highest popularity value.

3. The method for generating phishing practice emails according to claim 1, characterized in that, The email corpus features include topic structure distribution, body structure distribution, tone style distribution, business action distribution, and body length distribution. The step of generating email corpus feature information based on email corpus from the target person's organizational neighborhood includes: Sensitive entity identification and de-identification processing are performed on the email corpus, and feature information of the email corpus is extracted.

4. The method for generating phishing practice emails according to claim 1, characterized in that, The target constraint also includes revealing the sender's identity; Before generating target constraints based on the business scenario feature information and the email corpus feature information, the phishing exercise email generation method further includes: Obtain the email communication metadata of the target personnel, wherein the email communication metadata includes at least one of the following: sender, recipient, sending time, sender identity type, email subject category, and associated business system; Based on the email communication metadata, a communication relationship graph centered on the target person is constructed, and the communication relationship graph includes the communication relationship between the target person and at least one candidate sender identity; The overall matching degree of each candidate sender identity is determined based on the communication relationship diagram, and the sender identity to be displayed is determined from each candidate sender identity based on the overall matching degree.

5. The method for generating phishing practice emails according to claim 1, characterized in that, The multidimensional verification also includes writing structure verification and sensitive information verification; The process of performing multi-dimensional verification on the initial training email and outputting the verification results includes: The initial training emails are subjected to the following checks: similarity to real emails, business boundary and consistency, writing structure and sensitive information, generating multiple single-dimensional check results. If all the single-dimensional verification results indicate that the initial training email has passed the multi-dimensional verification, then the initial training email is determined to have passed the multi-dimensional verification; or If any single-dimensional verification result fails, the initial training email is determined to have failed the multi-dimensional verification.

6. The method for generating phishing practice emails according to claim 1, characterized in that, Also includes: If the verification result indicates that the initial training email has failed the multidimensional verification, the initial training email is marked as verification failed, and candidate training emails are regenerated based on the target constraints.

7. A device for generating phishing practice emails, characterized in that, include: The determination module is used to determine the business scenario feature information of the initial exercise email based on the intranet access events of the target personnel and preset evaluation indicators, and to generate email corpus feature information based on the email corpus of the target personnel's organizational neighborhood. The generation module is used to generate target constraints based on the business scenario feature information and the email corpus feature information. The target constraints include at least a list of allowed fields, a list of prohibited content rules, and writing constraints. The generation module is also used to generate initial exercise emails based on the target constraints. The verification module is used to perform multi-dimensional verification on the initial exercise email and output the verification results. The multi-dimensional verification includes at least real email similarity verification and business boundary and consistency verification. The determining module is further configured to determine the initial training email as a target phishing training email when the verification result indicates that the initial training email has passed the multi-dimensional verification.

8. An electronic device, characterized in that, It includes a processor and a memory, the memory storing a program or instructions that can run on the processor, the program or instructions being executed by the processor to implement the phishing exercise email generation method as described in any one of claims 1-6.

9. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a computer, implement the phishing exercise email generation method as described in any one of claims 1-6.

10. A computer program product, characterized in that, The computer program product is stored in a storage medium, and when executed by a computer, the computer program product implements the phishing exercise email generation method as described in any one of claims 1-6.