Penetration testing system and method based on AI large model and storage medium
Through a penetration testing system based on AI large-scale models, real-time monitoring and analysis of screen operation information, identifying security issues and generating reports, the problems of low output efficiency and complex results interpretation in existing penetration tests are solved, and the penetration test is automated and Intelligentization improves security management efficiency and decision-making quality.
Patent Information
- Application Number
- CN202411409879.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-10
- Publication Date
- 2024-12-03
AI Technical Summary
The existing penetration testing methods have problems such as low report output efficiency, complex result interpretation and lack of automation and intelligent support, resulting in inefficient security management and impact on the quality of decision-making.
A penetration testing system based on AI large model is adopted to monitor screen operation information in real time, extract features and analyze, identify security issues and generate correction suggestions, automatically fill penetration testing reports, and use pre-trained models to evaluate, score and report generation.
It significantly improves the efficiency and quality of penetration testing, automates the testing process, reduces the work burden of security experts, improves the readability of reports and management acceptance, and enhances network security protection capabilities.
Smart Images

Figure CN119065983A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of network penetration testing, and specifically relates to a penetration testing system, method and storage medium based on an AI big model. Background Art
[0002] Penetration testing is an active network security assessment method that simulates the behavior of hackers and conducts various attack attempts on the target system to discover and exploit vulnerabilities in the system and evaluate the security of the system. Vulnerability analysis refers to an in-depth analysis of the target system or software to discover existing security risks and defects, providing a basis for the next attack attempt; attack attempts refer to the implementation of various simulated attack methods on the target system during the penetration test to verify the system's defense capabilities, including the use of injection, brute force cracking, privilege escalation and other attack methods. Existing penetration testing solutions mainly include: ① Penetration testing tools based on script automation, such as Metasploit, Nmap, etc., which can partially realize automated testing, but still require a lot of manual participation and analysis; ② Penetration testing models based on machine learning, such as using deep learning to predict vulnerabilities and attack paths, but lack support for the entire testing process. Secondly, the existing penetration test report automatic generation technology based on natural language processing can improve the efficiency of report output, but still cannot solve the problem of difficult report understanding.
[0003] In summary, the prior art mainly has the following defects: ① Low report output efficiency: Since penetration test reports are mostly written manually, the report output speed is slow and the test results cannot be fed back quickly. This inefficiency will affect security managers' timely discovery and repair of system vulnerabilities.
[0004] ② Complex interpretation of results: Existing penetration test reports contain a lot of professional terms and complex details, which makes it difficult for non-technical personnel to understand the test results and reduces the acceptance of the test results by management. This problem of difficult-to-understand reports will affect the quality of security management decisions.
[0005] ③ Lack of automation and intelligent support: The existing methods are difficult to achieve automation and intelligent support for the entire process of penetration testing, requiring a lot of manual participation and analysis, and are inefficient. This lack of automation will increase the workload of security testers.
[0006] At present, there is an urgent need for a new method that can make full use of AI technology to comprehensively improve the efficiency and intelligence level of penetration testing. Summary of the invention
[0007] The purpose of the present invention is to provide a penetration testing system, method and storage medium based on an AI big model, aiming to solve the above-mentioned problems.
[0008] The present invention is mainly achieved through the following technical solutions: A penetration testing method based on an AI big model, comprising the following steps: Step S1: real-time monitoring and capturing of screen operation information during the penetration test; the screen operation information includes mouse movement trajectory, keyboard input, page element changes, and test tool behavior information; Step S2: Extract the features of screen operation information and input them into the trained AI model for analysis, identify possible security issues during the penetration test, and generate correction suggestions; Step S3: Evaluate and score the penetration test results based on the evaluation model; then, automatically fill in and generate a penetration test report based on the standard template.
[0009] In order to better implement the present invention, further, step S2 includes the following steps: Step S21: feature extraction: extracting features of screen operation information, and the features include mouse track features, keyboard input features and page element change features; then, proceed to step S22 and step S23 respectively; Step S22: Pattern recognition: Analyze the extracted features based on the trained AI big model. If an abnormal behavior pattern matching the known vulnerability pattern is identified, proceed to step S24; Step S23: Anomaly detection: Detect outliers in the data based on statistical methods, or identify abnormal operations based on the trained AI model. If potential abnormal operations are identified, proceed to step S24; Step S24: Based on the trained AI big model, the identified abnormal operation is compared with the vulnerability knowledge base, and possible security issues are output; Step S25: Generate corresponding correction suggestions based on the trained AI big model according to the identified vulnerability type.
[0010] In order to better implement the present invention, further, in step S21, the mouse trajectory features include identifying the total distance and movement rate of abnormally fast mouse movements and the concentration for identifying key areas of user attention, and the concentration is the time the mouse stays in a specific area; the keyboard input features include identifying the input frequency of key operations of high-frequency input and special characters for identifying possible attack behaviors; the page element change features include recording when and why DOM changes occur and event triggers that identify potential security vulnerabilities, and the event trigger is the relationship between page events and element changes.
[0011] In order to better implement the present invention, further, in the step S21, before feature extraction, the screen operation information is subjected to data preprocessing, and the data preprocessing includes data cleaning and data standardization. During data cleaning, irrelevant operations and noise data are removed; during data standardization, the mouse trajectory and keyboard input are converted into a unified time series format.
[0012] In order to better implement the present invention, further, in step S2, correction suggestions are generated based on rules: according to the identified vulnerability type, preset rules are applied to generate correction suggestions; or according to the running status and response results of the test tool, the test strategy and parameters are dynamically adjusted; or correction suggestions are generated based on the AI big model.
[0013] In order to better implement the present invention, further, in step S2, training the AI big model includes the following steps: Step A1: Build a pre-trained model of the AI large model based on the Transformer architecture; Step A2: Collect data sets: Collect penetration test related data sets, and the penetration test related data sets include historical penetration test reports, vulnerability descriptions, and repair suggestions, covering several attack vectors and defense strategies; Step A3: Train the pre-trained model based on the dataset: Training based on fine-tuning: fine-tune the pre-trained model on the dataset and optimize it for specific penetration testing tasks; or training based on supervised learning: use labeled data for training, and the annotations include vulnerability type, severity, and repair method; or training based on incremental learning: regularly update the model to improve its accuracy and adaptability by introducing new test data and feedback.
[0014] In order to better implement the present invention, further, step S3 includes the following steps: Step S31: designing a standard template for a penetration test report, the standard template including several chapter structures and format requirements; Step S32: Automatically fill in the contents of each chapter in the report template according to the result of step S2; Step S33: using the template engine or Word SDK technology to beautify the report format and improve the report readability; Step S34: Automatically generate key screenshots and log attachments during the test process.
[0015] In order to better implement the present invention, further, in step S3, the penetration test results are evaluated and scored based on severity, scope of impact and availability, and weights are set for severity, scope of impact and availability respectively.
[0016] The present invention is mainly achieved through the following technical solutions: A penetration testing system based on an AI big model, which is performed based on the above-mentioned penetration testing method based on an AI big model, includes an AI Eyes screen monitoring module, an AI analysis and correction module, a report generation module and a human-computer interaction module; The AI Eyes screen monitoring module is used to capture screen operation information during the immersion test in real time; The AI analysis and correction module is used to receive data collected by AI Eyes, analyze and judge using the pre-trained AI large model, and provide possible safety issues and correction suggestions during the test process; The report generation module is used to automatically generate a penetration test report based on the analysis results of the AI big model, and the report includes an overview of the test process, discovered vulnerabilities, and repair suggestions; The human-computer interaction module is used to provide a visual interface for testers to achieve control and interaction of various functional modules of the system.
[0017] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the above-mentioned penetration testing method based on the AI big model.
[0018] The beneficial effects of the present invention are as follows: The present invention uses the AI big model to solve the problems of report output and auxiliary penetration testing in existing penetration testing. The present invention uses AI Eyes to monitor the screen operation data and transmits the data to the AI big model. The AI big model analyzes and judges the data and outputs auxiliary results. At the same time, the AI big model can also automatically generate report content based on the data in the penetration test process. The present invention can effectively improve the efficiency and quality of penetration testing and provide intelligent support for security assessment.
[0019] The present invention makes full use of AI technology to automatically perform each link of penetration testing, greatly improving the testing efficiency. The AI system can automatically analyze the vulnerabilities of the target system, generate test plans, and automatically execute attack attempts, thereby reducing the workload of security experts and enabling penetration testing to be completed more quickly. The AI system of the present invention has the ability to continuously learn and update, can pay attention to security vulnerability intelligence in real time, dynamically adjust test strategies, and ensure the discovery of the latest system vulnerabilities. This is particularly important in the face of a rapidly changing network security threat environment. The present invention integrates the professional skills of penetration testing into the AI system, so that organizations with limited resources such as small and medium-sized enterprises can also use the AI system to conduct effective penetration testing and improve overall network security protection capabilities.
[0020] Integrating the professional skills of penetration testing into the AI system enables organizations with limited resources, such as small and medium-sized enterprises, to effectively conduct penetration testing. It lowers the technical threshold for penetration testing and improves the overall network security protection capabilities. The visual display of the penetration testing process and results makes it easier for security managers to intuitively understand and analyze the test situation. It provides targeted repair suggestions based on the test results to guide users to effectively repair system vulnerabilities. The system has cross-platform compatibility and can be deployed in different operating system environments. The system can be integrated with the company's existing security management platform to achieve unified management of functions. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] Figure 1 This is the operation flow chart of the penetration testing system based on the AI big model. DETAILED DESCRIPTION
[0022] Embodiment 1: A penetration testing method based on an AI big model specifically includes the following steps: Step 1: AI Eyes screen monitoring, mainly including the following key technologies: (1) Screen capture: Use Windows API or other screen capture tools to capture the screen in real time.
[0023] (2) Mouse track recording: Use Windows API to monitor mouse events, record mouse movement tracks, and analyze user operation habits and key areas.
[0024] (3) Keyboard input recording: Use Windows API to monitor keyboard events, record keyboard input, and analyze test ideas and key operations.
[0025] (4) Page element monitoring: Use DOM element monitoring technology to detect changes in key elements of the page in real time and discover key operations and results during the test.
[0026] (5) Test tool monitoring: Use the Windows process monitoring API to obtain the running status and response results of test tools (such as Burp Suite, Metasploit, etc.) in real time.
[0027] Step 2: AI analysis and correction, mainly including the following key technologies: (1) Problem identification: Use pre-trained AI models, such as BERT and GPT, combined with a vulnerability knowledge base, to analyze the captured data and identify possible problems during the testing process.
[0028] (2) Correction suggestions: Based on the problem identification results, corresponding correction suggestions are given using rule-based or generative model-based methods.
[0029] (3) Result evaluation: Use the pre-trained evaluation model to evaluate and score the severity and scope of impact of the problem, providing a basis for report generation.
[0030] Step 3: Report generation, mainly including the following key technologies: (1) Report template: Design a standard template for penetration testing reports, including the structure and format requirements of each chapter.
[0031] (2) Content filling: Automatically fill in the contents of each chapter in the report template based on the results of the AI analysis and correction module.
[0032] (3) Format beautification: Use technologies such as template engines or Word SDK to beautify the report format and improve report readability.
[0033] (4) Attachment generation: Automatically generate key screenshots, log records and other attachments during the test process.
[0034] Specifically, the penetration test report mainly includes the following contents: (1) Overview of the test process: including basic information such as test objectives, test methods, and test time.
[0035] (2) Discovered vulnerabilities: A detailed description of the discovered vulnerabilities, including vulnerability type, impact scope, and degree of harm.
[0036] (3) Repair suggestions: For the discovered vulnerabilities, corresponding repair suggestions are given, including repair methods, estimated repair costs, etc.
[0037] (4) Appendix: includes key screenshots, log records and other auxiliary information during the test process.
[0038] Preferably, the data processing in step 2 is as follows: (1) Input data type: Screen operation data (mouse trajectory, keyboard input, etc.); Page element change data (DOM changes); Test tool logs and response results; (2) Data preprocessing: Data cleaning: Remove invalid or duplicate operation records. Filter out irrelevant noise data, such as random clicks by users.
[0039] Data formatting: Convert data into a unified format (such as time series) to facilitate subsequent analysis and model input.
[0040] (3) Feature extraction: Extract key features from screen operations, such as mouse movement speed, click frequency, interaction time of specific elements, etc. Keyboard input character type, input mode, etc. Key features include: a. Mouse track characteristics: Distance and Speed: Calculates the total distance and speed of mouse movement, and identifies abnormally fast movements.
[0041] Concentration: Analyze the time the mouse stays in a specific area to identify the key areas of user attention.
[0042] b. Keyboard input features: Input frequency: Record the number of inputs within a specific period of time and identify key operations with high-frequency inputs.
[0043] Special character usage: Analyzes input special characters (such as carriage return, Tab key, etc.) to identify possible attack behaviors (such as SQL injection attempts).
[0044] c. Characteristics of page element changes: DOM changes: monitor changes to key elements (such as button states, text content, etc.), and record when and why they change.
[0045] Event triggering: Analyze the relationship between page events (such as clicks, hovers, etc.) and element changes to identify potential security vulnerabilities.
[0046] Contextual information: Combine page status and user actions to extract contextual information to better understand the operation intention.
[0047] Preferably, the problem identification in step 2 is as follows: (1) Feature input: The extracted features are input into the pre-trained AI model (such as BERT, GPT).
[0048] BERT: Used to understand context and semantics, and identify potential security issues through deep learning of input features.
[0049] GPT: Used for generative analysis to generate possible security issues and attack paths based on input features.
[0050] (2) Pattern recognition: Use AI models to train the extracted features and identify behaviors that match known vulnerability patterns.
[0051] (3) Anomaly detection: Use statistical methods (such as Z-score or IQR) to detect abnormal behavior and determine whether the operation is within the normal range. Or use AI models to identify normal and abnormal operations.
[0052] (4) Reasoning based on knowledge base: Based on abnormal behavior or abnormal patterns, the AI model compares the identified features with the vulnerability knowledge base (such as CVE, OWASP Top Ten) to identify potential vulnerabilities.
[0053] Preferably, the correction suggestion in step 2 is generated as follows: (1) Rule-based approach: Generate correction suggestions for identified problems using preset rules (such as coding standards and best practices). For example, for XSS vulnerabilities, it is recommended to properly escape user input. For SQL injection, it is recommended to use parameterized queries.
[0054] (2) Or a generative model-based approach: Use a trained generative model (such as GPT-3) to automatically generate correction suggestions, providing more natural language output and context-related feedback.
[0055] (3) Or dynamic adjustment suggestions: Dynamically adjust the test strategy and parameters according to the running status and response results of the test tool. For example, if there are continuous failures, you can adjust the attack vector or parameter settings. For the security issues found, it is recommended to use other test tools for further verification.
[0056] Preferably, the result evaluation in step 2 is as follows: (1) Input data: identified security issues and their related characteristics.
[0057] (2) Evaluation indicators: severity, scope of impact, exploitability, etc.
[0058] (3) Calculate the evaluation score: Quantitative Scoring: A scoring range is set for each indicator, and an overall score is calculated based on severity, scope of impact, and exploitability, which is used to assess how easy it is to exploit the vulnerability.
[0059] The severity rating is as follows: High: Can lead to complete system control, sensitive data leakage, etc.
[0060] Medium: May affect the normal function of the system. There is a certain risk but it is not immediately fatal.
[0061] Low: The impact on the system is small and only a potential risk.
[0062] The impact sphere is rated as follows: Widespread: Multiple systems or modules are affected, which may affect most users.
[0063] Local: affects only a specific module or a small number of users.
[0064] Single: only affects a single function, with minimal impact on users.
[0065] The availability rating is as follows: Easy to exploit: Attackers only need to follow simple steps to exploit.
[0066] Moderate: Requires some technical background or tools.
[0067] Difficult: Requires complex operations or specific conditions.
[0068] You can set weights for each evaluation indicator, for example: severity (40%), impact (30%), and exploitability (30%).
[0069] Based on the rating of each indicator (e.g. 1-5 points), the final score is: Total score = (severity score × 0.4) + (scope score × 0.3) + (exploitability score × 0.3) Total score = (severity score × 0.4) + (scope score × 0.3) + (exploitability score × 0.3) Expert review: The preliminary scores are adjusted based on the experience and judgment of domain experts to ensure the accuracy and practicality of the assessment.
[0070] Qualitative assessment: Building on the quantitative assessment, the expert team can adjust the score to take into account the complexity and specific context of the actual situation.
[0071] Scoring team: The scoring is conducted by an evaluation team composed of security experts, penetration testers, and technical personnel in related fields. Team members should have relevant certifications (such as CISSP, CEH, etc.) and rich practical experience to ensure the professionalism and accuracy of the scoring.
[0072] The scoring process is as follows: Initial Assessment: A preliminary score is given to each vulnerability by the penetration tester.
[0073] Expert review: The scoring results are submitted to the evaluation team for review and discussion by experts, and re-scoring is performed if necessary.
[0074] Consensus formation: Team members discuss each vulnerability and reach a consensus to ensure fair and objective scoring.
[0075] Integrate the evaluation results to provide quantitative and qualitative basis for subsequent reports.
[0076] Comprehensive score: The final score of each vulnerability is the comprehensive result of quantitative and qualitative evaluation. Taking into account the opinions of experts and actual conditions, the final score can be adjusted appropriately.
[0077] Report generation: The assessment results will be included in the penetration test report, including a detailed description of each vulnerability, a score, and corresponding remediation recommendations.
[0078] Embodiment 2: A penetration testing system based on an AI big model, comprising: AI Eyes screen monitoring module: Based on computer vision technology, it captures screen operation information during penetration testing in real time, including mouse tracks, keyboard input, page element changes, etc.
[0079] AI analysis and correction module: Receives data collected by AI Eyes, uses pre-trained AI large models to analyze and judge, and provides possible problems and correction suggestions during the test process.
[0080] Report generation module: Based on the AI analysis results, the penetration test report is automatically generated, including an overview of the test process, discovered vulnerabilities, and repair suggestions.
[0081] Human-computer interaction module: provides a visual interface for testers to realize control and interaction of various functional modules of the system.
[0082] Preferably, the specific implementation scheme of the AI analysis and correction module is as follows: 2.1. Methods for Pre-training Large AI Models (1) Model selection: Choose a pre-trained model suitable for penetration testing, such as models based on the Transformer architecture (such as BERT or GPT), because they perform well in processing natural language and context understanding.
[0083] (2) Dataset preparation: Collect large-scale penetration test-related datasets, including historical penetration test reports, vulnerability descriptions, and repair suggestions. Ensure that the dataset covers multiple attack vectors and defense strategies.
[0084] (3) Training is performed using any of the following methods: Fine-tuning: Fine-tune the pre-trained model on the collected dataset and optimize it for specific penetration testing tasks.
[0085] Supervised learning: Use labeled data for training, with annotations including vulnerability type, severity, and fix method.
[0086] Incremental learning: Update the model regularly to improve its accuracy and adaptability by introducing new test data and feedback.
[0087] 2.2. Specific methods of analysis and judgment 1) Data analysis process: Feature extraction: Extract features from the data collected by the AI Eyes screen monitoring module, such as mouse click location, keyboard input frequency, page loading time, etc.
[0088] Pattern recognition: Use machine learning algorithms (such as K-means, random forest, etc.) to analyze the extracted features and identify abnormal behavior patterns.
[0089] 2) Use any of the following methods to make a judgment: Rule-based judgment: Set a series of rules (for example, a large amount of input in a short period of time, repeated visits to specific pages, etc.) to identify potential security issues.
[0090] Predictive model: Use the trained AI model to predict the data and determine whether there are known vulnerabilities or security risks.
[0091] 2.3. Methods for identifying problems through data analysis (1) Problem identification: Anomaly detection: Detect outliers in the data using statistical-based methods (such as Z-score or IQR) to identify unusual operating patterns.
[0092] Vulnerability mapping: Compare the identified operations with known vulnerability libraries (such as the CVE database) to identify possible security issues.
[0093] (2) Suggestions for amendment: Automated suggestions: Based on the identified vulnerability type, the AI model automatically generates corresponding remediation suggestions, such as code modifications, configuration adjustments, etc.
[0094] Prioritization: Prioritize remediation suggestions based on vulnerability severity and impact, helping users focus on the most critical issues.
[0095] During use of the present invention, Figure 1 As shown, first, start the AI Eyes screen monitoring module to capture the screen operation information during the penetration test in real time. Then, input the captured data into the AI analysis and correction module, and use the pre-trained model to identify problems and make correction suggestions. According to the AI analysis results, the report generation module automatically generates a penetration test report. The tester views the analysis results through the human-computer interaction module, and can perform manual intervention and secondary confirmation when necessary. After the report output is completed, the system will automatically archive it for subsequent review and statistical analysis.
[0096] The present invention is applicable to various penetration testing scenarios, for example: Internal penetration testing in an enterprise: It can help the enterprise IT security team improve the efficiency and coverage of penetration testing and discover potential vulnerabilities. Third-party penetration testing services: It can provide intelligent testing assistance means for penetration testing service providers to improve service quality. Education and training fields: It can be applied to the penetration testing practice link in information security education and training to improve the learning effect of trainees.
[0097] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Any simple modification or equivalent change made to the above embodiment based on the technical essence of the present invention shall fall within the protection scope of the present invention.
Claims
1. A penetration testing method based on AI big model, characterized in that: The following steps are involved: Step S1: real-time monitoring and capturing of screen operation information during the penetration test; the screen operation information includes mouse movement trajectory, keyboard input, page element changes, and test tool behavior information; Step S2: Extract the features of screen operation information and input them into the trained AI model for analysis, identify possible security issues during the penetration test, and generate correction suggestions; Step S3: Evaluate and score the penetration test results based on the evaluation model; Then, based on the standard template, the penetration test report is automatically populated and generated.
2. According to claim 1, a penetration testing method based on an AI large model is characterized in that: The step S2 comprises the following steps: Step S21: feature extraction: extracting features of screen operation information, and the features include mouse track features, keyboard input features and page element change features; then, proceed to step S22 and step S23 respectively; Step S22: Pattern recognition: Analyze the extracted features based on the trained AI big model. If an abnormal behavior pattern matching the known vulnerability pattern is identified, proceed to step S24; Step S23: Anomaly detection: Detect outliers in the data based on statistical methods, or identify abnormal operations based on the trained AI model. If potential abnormal operations are identified, proceed to step S24; Step S24: Based on the trained AI big model, the identified abnormal operation is compared with the vulnerability knowledge base, and possible security issues are output; Step S25: Generate corresponding correction suggestions based on the trained AI big model according to the identified vulnerability type.
3. According to claim 2, a penetration testing method based on an AI large model is characterized in that: In step S21, the mouse trajectory features include the total distance and movement rate of the mouse for identifying abnormally fast movement and the concentration for identifying the key area of user attention, wherein the concentration is the residence time of the mouse in a specific area; The keyboard input features include identifying the input frequency of key operations with high-frequency input and special characters for identifying possible attack behaviors; the page element change features include recording DOM changes when and why changes occur and event triggers for identifying potential security vulnerabilities, and the event triggers are the relationship between page events and element changes.
4. According to claim 2, a penetration testing method based on an AI large model is characterized in that: In the step S21, before feature extraction, the screen operation information is subjected to data preprocessing, and the data preprocessing includes data cleaning and data standardization. During data cleaning, irrelevant operations and noise data are removed; during data standardization, the mouse trajectory and keyboard input are converted into a unified time series format.
5. According to claim 1, a penetration testing method based on an AI large model is characterized in that: In the step S2, a correction suggestion is generated based on the rule: according to the identified vulnerability type, a correction suggestion is generated by applying the preset rule; Or dynamically adjust test strategies and parameters according to the operating status and response results of the test tool; or generate correction suggestions based on the AI big model.
6. According to claim 1, a penetration testing method based on an AI large model is characterized in that: In step S2, training the AI big model includes the following steps: Step A1: Build a pre-trained model of the AI large model based on the Transformer architecture; Step A2: Collect data sets: Collect penetration test related data sets, and the penetration test related data sets include historical penetration test reports, vulnerability descriptions, and repair suggestions, covering several attack vectors and defense strategies; Step A3: Train the pre-trained model based on the dataset: Training based on fine-tuning: fine-tune the pre-trained model on the dataset and optimize it for specific penetration testing tasks; or training based on supervised learning: use labeled data for training, and the annotations include vulnerability type, severity, and repair method; or training based on incremental learning: regularly update the model to improve its accuracy and adaptability by introducing new test data and feedback.
7. The penetration testing method based on AI big model according to claim 1 is characterized in that: Step S3 includes the following steps: Step S31: designing a standard template for a penetration test report, the standard template including several chapter structures and format requirements; Step S32: Automatically fill in the contents of each chapter in the report template according to the result of step S2; Step S33: using the template engine or Word SDK technology to beautify the report format and improve the report readability; Step S34: Automatically generate key screenshots and log attachments during the test process.
8. According to claim 1, a penetration testing method based on an AI large model is characterized in that: In step S3, the penetration test results are evaluated and scored based on severity, impact scope and availability, and weights are set for severity, impact scope and availability respectively.
9. A penetration testing system based on an AI big model, based on the penetration testing method based on an AI big model according to any one of claims 1 to 8, characterized in that: Including AI Eyes screen monitoring module, AI analysis and correction module, report generation module and human-computer interaction module; The AI Eyes screen monitoring module is used to capture screen operation information during the immersion test in real time; The AI analysis and correction module is used to receive data collected by AI Eyes, analyze and judge using the pre-trained AI large model, and provide possible safety issues and correction suggestions during the test process; The report generation module is used to automatically generate a penetration test report based on the analysis results of the AI big model, and the report includes an overview of the test process, discovered vulnerabilities, and repair suggestions; The human-computer interaction module is used to provide a visual interface for testers to achieve control and interaction of various functional modules of the system.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the penetration testing method based on the AI big model described in any one of claims 1-8 is implemented.
Citation Information
Cited By
CPE identification system based on ChatGPT large model and incremental learning technology
CN120030549A
A CPE identification system based on ChatGPT large model and incremental learning technology
CN120030549B
ECPS test link execution method based on LLM and feedback improvement mechanism
CN122221247A
Business logic vulnerability automatic penetration test method based on RAG and multiple agents
CN122419825A
Intelligent model penetration testing method based on RAG
CN122471446A
