A method and related equipment for email threat defense
By employing a two-tiered analysis mechanism, combined with sending metadata and large language model analysis, the email gateway backlog problem was resolved, enabling efficient identification and filtering of malicious emails and improving the processing efficiency and security of the email system.
Patent Information
- Application Number
- CN202611093457.0
- 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
In existing technologies, text feature-based email detection methods cause severe backlogs in email gateways, making it impossible to efficiently filter malicious emails.
Through a two-tiered analysis mechanism, suspicious email sources are first identified based on the email's metadata and added to a dynamic risk list. Then, structured parsing and large language model analysis are performed, combined with a threat intelligence database and whitelist, to accurately determine whether the email is malicious.
It achieves efficient filtering of email gateways, reduces processing delays for legitimate emails, improves the accuracy of malicious email identification, and ensures network security.
Smart Images

Figure CN123122869A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of network security technology, specifically to a method, apparatus, electronic device, and computer-readable storage medium for email threat defense. Background Technology
[0002] With the deepening of enterprise informatization, email has become the main medium for daily business collaboration and core data flow. However, in recent years, malicious email attacks, such as phishing attacks, social engineering fraud, and credential theft, have occurred frequently, posing a serious threat to enterprise assets and network perimeter security.
[0003] Related technologies for detecting phishing emails based on text features typically involve a full scan of all email bodies and attachments according to a pre-defined rule-based thesaurus. When a keyword or rule is matched, an interception is triggered. However, performing an indiscriminate large-scale model scan on all emails can lead to severe backlogs at email gateways. Summary of the Invention
[0004] This invention provides an email threat defense method and related equipment, which achieves precise allocation of computing power through two-level analysis, quickly filters most normal emails, and solves the problem of severe backlog in email gateways.
[0005] In a first aspect, embodiments of the present invention provide a method for email threat defense. The method includes:
[0006] Collect sending metadata at the inbound receiving boundary of the recipient's email system, identify suspicious sending sources with abnormalities based on the sending metadata, and add suspicious sending sources to a dynamic risk list;
[0007] The system retrieves emails sent by suspicious sources and returns a delayed delivery response code to the email receiving gateway; it then performs structured parsing on the emails to obtain the parsing result data.
[0008] Based on the parsed data, prompt words are constructed and input into a large language model to generate semantic analysis results; the semantic analysis results are used to indicate the email intent.
[0009] Extract domain names from the links in the parsed data, match the domain names with a preset threat intelligence database to obtain associated threat tags; match the domain names with a preset whitelist, and remove the corresponding threat tags from trusted domain names that match the preset whitelist.
[0010] Based on the semantic analysis results and hazard tags, it is determined whether the email is malicious. If it is determined to be malicious, it is blocked; if it is determined to be non-malicious, it is allowed to be delivered.
[0011] In one implementation, the step of identifying suspicious sending sources with anomalies based on sending metadata includes:
[0012] Normalize the metadata of the sent emails;
[0013] The anomaly score is obtained by weighted summation of the normalized sending metadata according to preset weights.
[0014] If the abnormal score exceeds the preset risk threshold, the corresponding source is determined to be a suspicious source.
[0015] In one implementation, if a suspicious source is added to a dynamic risk list and is not identified as abnormal again within a preset time period, it is removed from the dynamic risk list.
[0016] In one implementation, constructing prompt words based on the parsed result data includes:
[0017] The system identifies images within the email body of the parsed data, obtains the text content carried by the images, merges the text content with the email body text based on the context, and performs text cleaning on the merged text to obtain the fused text.
[0018] Prompts are constructed based on predefined dual-agent adversarial CoT prompt templates, fused text, and email subject lines within the parsed data.
[0019] In one embodiment, the method further includes:
[0020] If the email is determined to be malicious, a silent cleanup command is sent to the recipient's email system, domain blocking rules are written to the border firewall and email security gateway, and EDR is coordinated to trace the terminals that have accessed the domain and perform network isolation.
[0021] In one implementation, the dynamic risk list uses a Redis instance as its storage medium.
[0022] Secondly, embodiments of the present invention provide an email threat defense device. The device includes:
[0023] The email metadata anomaly profiling module is used to collect sending metadata at the inbound receiving boundary of the recipient's email system, identify suspicious sending sources with anomalies based on the sending metadata, and add suspicious sending sources to a dynamic risk list;
[0024] The parsing module is used to obtain emails sent by suspicious sources and return a delayed delivery response code to the email receiving gateway; it performs structured parsing on the emails to obtain parsing result data.
[0025] The secondary analysis module is used to construct prompt words based on the parsed data, input them into a large language model, and generate semantic analysis results; the semantic analysis results are used to indicate the email intent.
[0026] The cascading verification module is used to extract domain names from the links in the parsed result data, match the domain names with a preset threat intelligence database to obtain associated harm tags; match the domain names with a preset whitelist, and remove the corresponding harm tags from trusted domain names that match the preset whitelist.
[0027] The decision module is used to determine whether the email is malicious based on the semantic analysis results and the harm tag. If it is determined to be a malicious email, it will be blocked; if it is determined to be a non-malicious email, it will be allowed to be delivered.
[0028] In one implementation, the email metadata anomaly profiling module includes:
[0029] The normalization submodule is used to normalize the sent data metadata.
[0030] The calculation submodule is used to perform a weighted summation of the normalized sending metadata according to preset weights to obtain the anomaly score.
[0031] The judgment submodule is used to determine the corresponding sending source as a suspicious sending source if the abnormal score exceeds the preset risk threshold.
[0032] In one implementation, the email metadata anomaly profiling module is further used to remove the suspicious email source from the dynamic risk list if it is not identified as having anomalies again within a preset time period after being added to the dynamic risk list.
[0033] In one implementation, the secondary analysis module includes:
[0034] The text processing submodule is used to identify images in the email body within the parsed data, obtain the text content carried by the images, merge the text content with the email body text in combination with the context, and perform text cleaning on the merged text to obtain the fused text.
[0035] The prompt word construction submodule is used to construct prompt words based on predefined dual-agent adversarial CoT prompt word templates, fused text, and email subject within the parsed result data.
[0036] In one implementation, the decision module is further configured to, if determined to be a malicious email, issue a silent cleanup command to the recipient's email system, write domain name blocking rules to the border firewall and email security gateway, and coordinate with EDR to trace the terminals that have accessed the domain name and perform network isolation.
[0037] In one implementation, the dynamic risk list uses a Redis instance as its storage medium.
[0038] Thirdly, the present invention provides an electronic device, the electronic device comprising:
[0039] At least one processor; and
[0040] A memory communicatively connected to the at least one processor; wherein,
[0041] The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the email threat defense method described above.
[0042] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions that are used to cause a processor to execute the above-described email threat defense method.
[0043] In this embodiment of the invention, precise allocation of computing power is achieved through a two-stage analysis. The first stage identifies suspicious sending sources with abnormalities based on sending metadata and adds them to a dynamic risk list to quickly filter out most legitimate sending sources. The second stage analyzes only suspicious emails that hit the sending sources in the dynamic risk list, ensuring deep detection capabilities while keeping computing power consumption within an acceptable engineering range and resolving the problem of severe backlog in email gateways. Attached Figure Description
[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0045] Figure 1 This is a flowchart of an email threat defense method provided in Embodiment 1 of the present invention;
[0046] Figure 2 This is a schematic diagram of the structure of an email threat defense device provided in Embodiment 2 of the present invention;
[0047] Figure 3 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation
[0048] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.
[0049] Figure 1 This is a flowchart of an email threat defense method provided in Embodiment 1 of the present invention. This embodiment can be applied to email threat defense devices, such as... Figure 1 As shown, this embodiment may include the following steps:
[0050] Step 101: Collect sending metadata at the inbound receiving boundary of the recipient's email system, identify suspicious sending sources with abnormalities based on the sending metadata, and add the suspicious sending sources to the dynamic risk list.
[0051] The email threat defense device's mailbox metadata anomaly profiling module can be mounted on the MTA (Mail Transfer Agent) via the Milter protocol, collecting sending metadata for each inbound email at the inbound receiving boundary of the recipient's email system (such as Coremail). This access method requires no code intrusion or functional modification to any processing flow of the email system, significantly reducing deployment and maintenance costs.
[0052] The sending metadata may include: protocol metadata, such as sending IP, sending domain name, HELO / EHLO host identifier, SPF verification result, DKIM signature verification status and DMARC domain name alignment verification result, etc., as well as behavioral statistics metadata, such as the sending frequency of the corresponding sending IP within a 24-hour sliding window, session retry failure rate and SMTP interaction command sequence anomaly score, etc.
[0053] After collecting the sending metadata, suspicious sending sources with abnormalities are identified based on the sending metadata. The specific process can be as follows:
[0054] Sub-step 1011: Normalize the sending metadata.
[0055] For example, risk binary mapping is performed on protocol metadata such as SPF verification results, DKIM signature verification status, and DMARC domain name alignment verification results, with normal = 0 and abnormal / verification failure = 1; the sending frequency, session retry failure rate, and SMTP interaction command sequence abnormal score of the corresponding sending IP within the 24-hour sliding window are processed by min-max normalization, and the indicators are uniformly mapped to the [0,1] interval, and positive risk alignment is performed, with higher values in the interval representing higher risks.
[0056] Sub-step 1012: The normalized sending metadata is weighted and summed according to the preset weights to obtain the anomaly score; if the anomaly score exceeds the preset risk threshold, the corresponding sending source is determined to be a suspicious sending source.
[0057] The preset weights can be assigned by experts based on experience, with a total weight sum of 1. Each normalized feature value is multiplied by its corresponding weight and then summed to obtain the anomaly score of the sending source. If the anomaly score exceeds a preset risk threshold (e.g., 0.4), the sending source is marked as a suspicious sending source and added to the dynamic risk list hosted by the Redis instance. The sending source can be a sending email address, sending IP address, or sending domain name.
[0058] Once a suspicious source is added to the dynamic risk list, it will be removed from the list if it is not identified as abnormal again within a preset time period. For example,
[0059] If a source is marked as suspicious and is not identified as abnormal again within 12 consecutive hours, the source record will be automatically removed from the dynamic risk list. This time-sensitive automatic cleanup mechanism can adapt to IP rotation scenarios such as operator dynamic IPs and enterprise shared proxy IPs, effectively avoiding false blocking issues caused by IP reuse.
[0060] If none of the following criteria are met: HELO / EHLO host identifier validity, SPF verification result, DKIM signature verification status and DMARC domain name alignment verification result, sending frequency of the corresponding sending IP in the sliding window, session retry failure rate and SMTP interaction command sequence anomaly score, or if the anomaly score is lower than the preset risk threshold, then delivery will be allowed directly.
[0061] Step 102: Obtain emails sent by suspicious sources and return a delayed delivery response code to the email receiving gateway; perform structured parsing on the emails to obtain the parsing result data.
[0062] In this step, emails sent by suspicious sources can be cloned to a sandbox, where they are structured and parsed. At the same time, a delayed delivery response code is returned to the mail receiving gateway to ensure that the consistency of the original delivery transactions of the mail system is not disrupted while the suspicious emails enter the security analysis channel.
[0063] The specific process of performing structured parsing on the email may include: splitting the email header and email body; extracting the email subject from the header; segmenting the email body; extracting the complete HTML source code; identifying explicit and hidden links within the HTML through DOM traversal; separating embedded images, independent image attachments, and other types of attachments and extracting file resources; finally, integrating all extracted information to generate standardized parsing result data; and destroying the original email cache in the sandbox after parsing to eliminate security remnants.
[0064] Step 103: Construct prompt words based on the parsing results data, and input them into the large language model to generate semantic analysis results; the semantic analysis results are used to indicate the email intent.
[0065] For emails where the body contains text presented as images (an attacking technique often used to circumvent text keyword filtering), the first step is to identify the images in the parsed data to obtain the text content carried by the images. Then, the text content is merged with the email body text in conjunction with the context. Finally, the merged text is cleaned, such as by performing Unicode obfuscation cleanup, to obtain the merged text.
[0066] Then, prompt words are constructed based on predefined dual-agent adversarial CoT prompt word templates, fused text, and email subject within the parsed data, and input into the large language model.
[0067] A large language model performs bidirectional adversarial logic simulation (Chain-of-Thought, CoT): for example, it simultaneously simulates reasoning adversarial scenarios from the perspectives of a "social engineering hacker" and a "security audit expert" to determine whether the email message contains phishing intentions such as emotional pressure inducement (e.g., account freezing forcing password changes), credential harvesting, or identity spoofing (e.g., impersonating HR to send a salary increase notification). Then, based on the risk scores corresponding to various fraud characteristics such as emotional pressure inducement, credential harvesting, identity spoofing, Unicode obfuscation, and hidden link inducement (each score ∈ [0,1]), a malicious score is obtained by performing an aggregation operation on a set of risk scores. Finally, a semantic analysis result containing the malicious score and a logical judgment report is output.
[0068] Step 104: Extract domain names from the links in the parsed result data, match the domain names with the preset threat intelligence database to obtain the associated harm tags; match the domain names with the preset whitelist, and remove the corresponding harm tags from trusted domain names that match the preset whitelist.
[0069] The preset threat intelligence database can refer to a database pre-configured locally that stores several known malicious domain names and corresponding threat tags, or it can be a database provided by a third party, such as the open-source collaborative threat intelligence community AlienVault Open Threat Exchange (OTX).
[0070] For example, after extracting the domain name from the links in the parsed result data, the OTX API is used to query its harm label (Phishing, Malicious, Spam, Active_C2, etc.) in OTX, and the query result is aligned and matched with the preset whitelist maintained locally. The corresponding harm label is removed from the trusted domain name that matches the preset whitelist.
[0071] Removing the corresponding threat tags from trusted domains that hit the preset whitelist can avoid the historical false alarm problem of preset threat intelligence databases, especially third-party intelligence databases; it also ensures that the local trusted domain policy takes priority over preset threat intelligence, such as external third-party intelligence, to prevent legitimate domains from being continuously marked as risky due to historical malicious records, thus preventing false blocking caused by raising the overall risk of emails.
[0072] Step 105: Based on the semantic analysis results and the hazard label, determine whether the email is a malicious email. If it is determined to be a malicious email, block it; if it is determined to be a non-malicious email, allow it to be delivered.
[0073] After obtaining the semantic analysis results and hazard tags, the system determines whether the email is malicious based on these results and tags. For example, firstly, an intelligence threat level score is calculated based on the hazard tags. The specific process can be as follows: A base score within the range [0,1] is pre-assigned to each hazard tag according to its severity; a higher score indicates a higher threat level. If the same domain matches multiple tags, the maximum base score is selected as the intelligence threat level score. If there are no hazard tags, the intelligence threat level score is 0. Then, a cascaded weighted calculation is performed on the maliciousness score and the intelligence threat level score to obtain the email composite risk assessment index. The specific process can be as follows: Corresponding weights are pre-configured for the maliciousness score and the intelligence threat level score. The score fusion calculation is performed according to a preset hierarchical order, first introducing the maliciousness score as the base risk score, then introducing the intelligence threat level score for weighted correction, ultimately generating the email composite risk assessment index. This index is used to comprehensively characterize the overall malicious risk level of the email. If the composite risk assessment index of an email is greater than the high-risk threshold, it is judged as a malicious email and blocked; if the composite risk assessment index of an email is less than the high-risk threshold, it is judged as a non-malicious email and allowed to be delivered.
[0074] The method may further include:
[0075] Step 106: If the email is determined to be malicious, a silent cleanup command is sent to the recipient's email system, domain name blocking rules are written to the border firewall and email security gateway, and EDR is linked to trace the terminals that have accessed the domain name and perform network isolation.
[0076] In this step, when the current email is determined to be malicious, the system initiates a comprehensive, closed-loop response mechanism. The specific execution process is as follows: First, a silent cleanup command is issued to the recipient's email system. Without the user's awareness, the malicious email is recycled, deleted, isolated, and archived in the background, preventing the user from viewing, clicking, or downloading the malicious content within the email, thus blocking the spread of risk from the email receiving link. Second, the high-risk domain name in the email is extracted and simultaneously written into the dynamic blocking rule base of the boundary firewall and email security gateway. A domain name access blocking policy is generated in real time to block all subsequent access requests to the malicious domain name from all terminals on the internal network, achieving boundary traffic interception. Finally, the system collaborates with the terminal EDR security platform to conduct a backtracking investigation, tracing in batches the terminal devices on the internal network that have accessed the malicious domain name. Temporary network isolation operations are automatically performed on the terminals involved, cutting off their communication links between the internal and external networks, preventing secondary security risks such as remote control by Trojans, lateral penetration, and data leakage, thus completing a full-link closed-loop defense from the email side, network side, to the terminal side.
[0077] This embodiment achieves precise allocation of computing power through a two-stage analysis. The first stage identifies suspicious sending sources with abnormalities based on sending metadata and adds them to a dynamic risk list to quickly filter out most legitimate sending sources. The second stage analyzes only suspicious emails that hit the sending sources in the dynamic risk list, ensuring deep detection capabilities while keeping computing power consumption within an acceptable engineering range and resolving the problem of severe backlog in email gateways.
[0078] By reconstructing the attacker's method of embedding text in images to circumvent keyword filtering, and combining a large language model to simultaneously simulate the dual perspectives of "attacker" and "security audit expert" in Chain-of-Thought reasoning adversarial, it can accurately identify phishing intentions such as emotional pressure inducement, identity spoofing, and credential theft in social engineering rhetoric, and has a significantly higher detection capability for cold domain phishing and image obfuscation phishing.
[0079] The method uses semantic analysis results and hazard tags to determine whether an email is malicious. The two signals complement each other: known threats already included in the intelligence are quickly identified, while new cold domain threats not covered by the intelligence are covered by semantic reasoning from a large model. The overall detection accuracy and robustness are significantly better than single-signal decision-making.
[0080] Once the current email is determined to be malicious, it triggers asynchronous parallel processing via the email system, firewall, and EDR, forming a complete closed-loop defense across the entire link from the email side, network side, to the terminal side.
[0081] Corresponding to the email threat defense method in this invention, this invention also provides an email threat defense device. Figure 2 This is a schematic diagram of the structure of an email threat defense device. Figure 2 As shown, the email threat defense device includes:
[0082] The email metadata anomaly profiling module 201 is used to collect sending metadata at the inbound receiving boundary of the recipient's email system, identify suspicious sending sources with anomalies based on the sending metadata, and add suspicious sending sources to a dynamic risk list;
[0083] The parsing module 202 is used to obtain emails sent by suspicious sources and return a delayed delivery response code to the email receiving gateway; it performs structured parsing on the emails to obtain parsing result data.
[0084] The secondary analysis module 203 is used to construct prompt words based on the parsing result data, input them into a large language model, and generate semantic analysis results; the semantic analysis results are used to indicate the email intent.
[0085] The cascading verification module 204 is used to extract domain names from the links in the parsed result data, match the domain names with a preset threat intelligence database to obtain associated harm tags; match the domain names with a preset whitelist, and remove the corresponding harm tags from trusted domain names that match the preset whitelist.
[0086] The decision module 205 is used to determine whether the email is malicious based on the semantic analysis results and the harm tag. If it is determined to be a malicious email, it will be blocked; if it is determined to be a non-malicious email, it will be allowed to be delivered.
[0087] In one embodiment, the email metadata anomaly profiling module 201 includes:
[0088] The normalization submodule is used to normalize the sent data metadata.
[0089] The calculation submodule is used to perform a weighted summation of the normalized sending metadata according to preset weights to obtain the anomaly score.
[0090] The judgment submodule is used to determine the corresponding sending source as a suspicious sending source if the abnormal score exceeds the preset risk threshold.
[0091] In one implementation, the email metadata anomaly profiling module 201 is further used to remove the suspicious email source from the dynamic risk list if it is not identified as having anomalies again within a preset time period after being added to the dynamic risk list.
[0092] In one embodiment, the secondary analysis module 203 includes:
[0093] The text processing submodule is used to identify images in the email body within the parsed data, obtain the text content carried by the images, merge the text content with the email body text in combination with the context, and perform text cleaning on the merged text to obtain the fused text.
[0094] The prompt word construction submodule is used to construct prompt words based on predefined dual-agent adversarial CoT prompt word templates, fused text, and email subject within the parsed result data.
[0095] In one implementation, the decision module 205 is further configured to, if determined to be a malicious email, issue a silent cleanup instruction to the recipient's email system, write domain name blocking rules to the border firewall and email security gateway, and coordinate with EDR to trace the terminals that have accessed the domain name and perform network isolation.
[0096] In one implementation, the dynamic risk list uses a Redis instance as its storage medium.
[0097] The email threat defense device provided in this embodiment of the invention can execute an email threat defense method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method execution.
[0098] Figure 3 A schematic diagram of an electronic device 30 that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.
[0099] like Figure 3 As shown, the electronic device 30 includes at least one processor 31 and a memory, such as a read-only memory (ROM) 32 or a random access memory (RAM) 33, communicatively connected to the at least one processor 31. The memory stores computer programs executable by the at least one processor. The processor 31 can perform various appropriate actions and processes based on the computer program stored in the ROM 32 or loaded from storage unit 38 into the RAM 33. The RAM 33 can also store various programs and data required for the operation of the electronic device 30. The processor 31, ROM 32, and RAM 33 are interconnected via a bus 34. An input / output (I / O) interface 35 is also connected to the bus 34.
[0100] Multiple components in electronic device 30 are connected to I / O interface 35, including: input unit 36, such as keyboard, mouse, etc.; output unit 37, such as various types of monitors, speakers, etc.; storage unit 38, such as disk, optical disk, etc.; and communication unit 39, such as network card, modem, wireless transceiver, etc. Communication unit 39 allows electronic device 30 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.
[0101] Processor 31 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 31 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 31 performs the various methods and processes described above, such as email threat defense methods.
[0102] In some embodiments, the email threat defense method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 38. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 30 via ROM 32 and / or communication unit 39. When the computer program is loaded into RAM 33 and executed by processor 31, one or more steps of the email threat defense method described above may be performed. Alternatively, in other embodiments, processor 31 may be used to execute the email threat defense method by any other suitable means (e.g., by means of firmware).
[0103] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include: implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.
[0104] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.
[0105] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.
[0106] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).
[0107] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.
[0108] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.
[0109] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.
[0110] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.
Claims
1. A method for defending against email threats, characterized in that, include: Collect sending metadata at the inbound receiving boundary of the recipient's email system, identify suspicious sending sources with abnormalities based on the sending metadata, and add suspicious sending sources to a dynamic risk list; Retrieve emails sent by suspicious sources and return a delayed delivery response code to the email receiving gateway; The email is parsed in a structured manner to obtain the parsing result data; Based on the parsed data, prompt words are constructed and input into a large language model to generate semantic analysis results; The semantic analysis results are used to indicate the intent of the email; Extract domain names from the links in the parsed data, match the domain names with a preset threat intelligence database, and obtain associated hazard tags; The domain name is matched with a preset whitelist, and the corresponding harm label is removed from the trusted domain name that matches the preset whitelist. Based on the semantic analysis results and hazard tags, it is determined whether the email is malicious. If it is determined to be malicious, it is blocked; if it is determined to be non-malicious, it is allowed to be delivered.
2. The method according to claim 1, characterized in that, The identification of suspicious sending sources with anomalies based on sending metadata includes: Normalize the metadata of the sent emails; The anomaly score is obtained by weighted summation of the normalized sending metadata according to preset weights. If the abnormal score exceeds the preset risk threshold, the corresponding source is determined to be a suspicious source.
3. The method according to claim 1, characterized in that: Once a suspicious source is added to the dynamic risk list, it will be removed from the list if it is not identified as abnormal again within a preset time period.
4. The method according to claim 1, characterized in that, The construction of prompt words based on the parsed data includes: The system identifies images within the email body of the parsed data, obtains the text content carried by the images, merges the text content with the email body text based on the context, and performs text cleaning on the merged text to obtain the fused text. Prompts are constructed based on predefined dual-agent adversarial CoT prompt templates, fused text, and email subject lines within the parsed data.
5. The method according to claim 1, characterized in that, Also includes: If the email is determined to be malicious, a silent cleanup command is sent to the recipient's email system, domain blocking rules are written to the border firewall and email security gateway, and EDR is coordinated to trace the terminals that have accessed the domain and perform network isolation.
6. The method according to any one of claims 1 to 5, characterized in that: The dynamic risk list uses a Redis instance as its storage medium.
7. An email threat defense device, characterized in that, include: The email metadata anomaly profiling module is used to collect sending metadata at the inbound receiving boundary of the recipient's email system, identify suspicious sending sources with anomalies based on the sending metadata, and add suspicious sending sources to a dynamic risk list; The parsing module is used to obtain emails sent by suspicious sources and return a delayed delivery response code to the email receiving gateway; The email is parsed in a structured manner to obtain the parsing result data; The secondary analysis module is used to construct prompt words based on the parsing results data, input them into the large language model, and generate word meaning analysis results; The semantic analysis results are used to indicate the intent of the email; The cascading verification module is used to extract domain names from the links in the parsed result data, match the domain names with a preset threat intelligence database, and obtain the associated hazard tags; The domain name is matched with a preset whitelist, and the corresponding harm label is removed from the trusted domain name that matches the preset whitelist. The decision module is used to determine whether the email is malicious based on the semantic analysis results and the harm tag. If it is determined to be a malicious email, it will be blocked; if it is determined to be a non-malicious email, it will be allowed to be delivered.
8. The apparatus according to claim 7, characterized in that, The secondary analysis module includes: The text processing submodule is used to identify images in the email body within the parsed data, obtain the text content carried by the images, merge the text content with the email body text in combination with the context, and perform text cleaning on the merged text to obtain the fused text. The prompt word construction submodule is used to construct prompt words based on predefined dual-agent adversarial CoT prompt word templates, fused text, and email subject within the parsed result data.
9. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the email threat defense method according to any one of claims 1-6.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the email threat defense method according to any one of claims 1-6.
