Data security protection system and method based on anti-ai penetration technology

Abstract

The application relates to the field of data security and network security, and discloses a data security protection system and method based on anti-AI penetration technology, which comprises the following steps: S1, receiving user identity data sent by a client; S2, generating a unique ePASS-ID through a national secret algorithm or other encryption algorithms; S3, according to a preset organizational structure information database, analyzing a digital appointment file to generate permission data corresponding to the ePASS-ID; S4, receiving a registration request of an AI entity, acquiring model fingerprint data of the AI entity, and generating an AI identity through a national secret algorithm or other encryption algorithms; S5, judging whether there is an anomaly based on a deep learning algorithm; and S6, processing business data. In the application, through the introduction of an identity pre-verification mechanism during the registration of an AI entity, an AI behavior white list rule and a real-time monitoring mechanism, the limitation that a legitimate AI cannot be distinguished from a malicious AI in the prior art is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data security and network security, and particularly relates to a data security protection system and method based on anti-AI penetration technology. BACKGROUND

[0002] In the prior art, data security protection systems usually protect data resources through access control lists (ACLs), firewall rules, and static identity authentication mechanisms. For example, role-based access control (RBAC) technology limits data access by predefining user permissions, and combines symmetric encryption algorithms (such as AES) or asymmetric encryption algorithms (such as RSA) to protect transmitted and stored data. In addition, some systems use log recording and rule matching methods to monitor abnormal behavior, such as detecting potential malicious access by setting API call frequency thresholds. These technologies have been widely applied in traditional network security scenarios and have achieved some protection against known threats.

[0003] In the prior art, AI entity identity spoofing attack methods include but are not limited to simulating user behavior through a generative adversarial network (GAN), spoofing biometric data, or probing system vulnerabilities through high-frequency API calls. These attacks usually occur in unregistered or anonymous access scenarios, and existing systems lack the ability to pre-validate AI entity identities and extract behavior characteristics. For example, traditional systems have difficulty distinguishing between legitimate AI (such as enterprise-deployed intelligent customer service) and malicious AI (such as automated scripts from external attackers), resulting in insufficient targeting of protective measures. SUMMARY

[0004] To address the above problems, the present application provides a data security protection system and method based on anti-AI penetration technology, aiming to improve the problem that traditional systems have difficulty distinguishing between legitimate AI (such as enterprise-deployed intelligent customer service) and malicious AI (such as automated scripts from external attackers).

[0005] In a first aspect, the present application provides a data security protection method based on anti-AI penetration technology, applied to a server side, comprising the following steps:

[0006] S1, receiving user identity data sent by a client, the user identity data including biometric data and behavior data;

[0007] S2, based on the user identity data, generating a unique ePASS-ID through a national encryption algorithm or other encryption algorithm, and encrypting and storing the biometric data using the national encryption algorithm or other encryption algorithm;

[0008] S3, according to the preset organizational structure information database, parsing the digital appointment file, generating the permission data corresponding to the ePASS-ID, the permission data is dynamically allocated based on a four-dimensional permission control model;

[0009] S4, receiving the registration request of the AI entity, pre-verifying the identity of the AI entity, obtaining the model fingerprint data after verification, generating the AI identity mark through the national secret algorithm or other encryption algorithm, and limiting the access range of the AI entity according to the AI behavior white list rule;

[0010] S5, real-time monitoring the API calling behavior of the AI entity, judging whether there is an exception based on a deep learning algorithm, if there is an exception, blocking the access of the AI entity and triggering an alarm;

[0011] S6, processing the business data, encrypting the transmission by using the national secret algorithm or other encryption algorithm, encrypting the storage by using the national secret algorithm or other encryption algorithm, and verifying the data integrity by using the national secret algorithm or other encryption algorithm.

[0012] Preferably, the biological feature data in the step S1 includes face, iris, voiceprint and finger vein data, the behavior data includes user operation habit data, and the user identity data is verified by a multi-modal certificate system.

[0013] Preferably, the four-dimensional permission control model in the step S3 includes:

[0014] The organizational structure dimension generates the permission inheritance relationship according to the enterprise hierarchical structure;

[0015] The data sensitivity dimension is based on the national secret algorithm or other encryption algorithm for hierarchical encryption of data;

[0016] The business scenario dimension uses the national secret algorithm or other encryption algorithm to encrypt the sandbox to isolate different business environments;

[0017] The time-space dimension limits the access permission according to the GPS and IP address.

[0018] Preferably, the AI behavior white list rule in the step S4 includes:

[0019] Registering the model fingerprint data of the AI entity and limiting the data interaction range thereof;

[0020] Prohibiting the AI entity from accessing sensitive data beyond the permission corresponding to the ePASS-ID thereof.

[0021] Preferably, the real-time monitoring in the step S5 includes:

[0022] Isolating the operation of the AI entity in the AI behavior sandbox;

[0023] If an abnormal API call is detected, a secondary biometric verification is triggered.

[0024] Preferably, the step S6 further comprises:

[0025] A national encryption algorithm or other encryption algorithm hash digital watermark is added to the exported data.

[0026] A multi-level approval process is set for sensitive data operations and is stored on a chain.

[0027] In a second aspect, the present application provides the following technical solution: a data security protection system based on anti-AI penetration technology, configured on a server side, comprising:

[0028] A receiving module is configured to receive user identity data and an AI entity registration request sent by a client;

[0029] An identity generation module is configured to generate an ePASS-ID based on the user identity data and generate an identity identifier for the AI entity;

[0030] A permission allocation module is configured to generate and allocate permission data according to an organizational structure information database;

[0031] A monitoring module is configured to monitor API call behavior of the AI entity in real time and determine abnormalities;

[0032] A data processing module is configured to encrypt and transmit and store business data.

[0033] Preferably, the identity generation module generates the ePASS-ID by using a national encryption algorithm or other encryption algorithm and encrypts biometric data by using the national encryption algorithm or other encryption algorithm.

[0034] In a third aspect, the present application provides the following technical solution: a computer device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned data security protection method based on anti-AI penetration technology.

[0035] In a fourth aspect, the present application provides the following technical solution: a readable storage medium, wherein the readable storage medium stores a computer program, and the computer program is executed by a processor to implement the above-mentioned data security protection method based on anti-AI penetration technology.

[0036] The present application has the following beneficial effects:

[0037] 1、In the present application, by introducing the identity pre-verification mechanism of AI entity registration, AI behavior white list rules and real-time monitoring mechanism, the fake identity and abnormal API calling behavior of AI entity (such as malicious generation model) can be effectively identified and blocked, and the legal AI is assigned a traceable identity label through digital signature and behavior log verification, and the behavior characteristic extraction model fingerprint data is generated, which solves the limitation that the existing technology cannot distinguish between legal and malicious AI.

[0038] 2、In the present application, the four-dimensional permission control model (organization structure, business scenario, data sensitivity, time-space) is proposed, which realizes the dynamic allocation and real-time update of permissions through data logic and database analysis technology, overcomes the limitation of single-dimensional static configuration in traditional permission management, can quickly adapt to the permission requirements in complex organization structure and multiple business scenarios, improves the adaptability of the system to diversified access control requirements, and reduces the artificial maintenance cost of permission management.

[0039] 3、In the present application, a multi-modal certificate system is adopted, biological feature data (face, iris, voiceprint, finger vein) and behavior data are combined, ePASS-ID is generated through national encryption algorithm or other encryption algorithm, and national encryption algorithm or other encryption algorithm is used, respectively, to realize asymmetric encryption transmission, symmetric encryption storage and integrity verification, compared with the traditional single encryption mode, the technical scheme significantly improves the anti-theft and anti-tamper ability of business data through multi-level encryption and verification mechanism, especially in the cross-network transmission and distributed storage environment, the security of data throughout the life cycle can be ensured.

[0040] 4、In the present application, by generating a unique identity label for AI entity and combining behavior sandbox isolation technology, real-time monitoring of AI operation and rapid positioning of abnormal behavior are realized, compared with the traditional indiscriminate management access control method, the technical scheme can record the model fingerprint and API calling log of AI entity, which is convenient for post-audit and responsibility tracing, so as to improve the management efficiency and security of the system in the scene of wide application of AI. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 The method flow chart of the data security protection method based on anti-AI penetration technology proposed in the present application;

[0042] Figure 2 The system architecture diagram of the data security protection system based on anti-AI penetration technology proposed in the present application. DETAILED DESCRIPTION

[0043] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all the other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0044] Embodiment one

[0045] With reference to Figure 1 In the first embodiment of the present application, the present application provides a data security protection method based on anti-AI penetration technology, applied to a server side, including the following steps:

[0046] S1, receiving user identity data sent by a client, the user identity data including biological feature data and behavior data;

[0047] S2, based on the user identity data, generating a unique ePASS-ID through a national secret algorithm or other encryption algorithm, and encrypting and storing the biological feature data by using the national secret algorithm or other encryption algorithm;

[0048] S3, according to a preset organizational structure information database, parsing a digital appointment file to generate permission data corresponding to the ePASS-ID, the permission data being dynamically allocated based on a four-dimensional permission control model;

[0049] S4, receiving a registration request of an AI entity, pre-verifying the identity of the AI entity, obtaining model fingerprint data of the AI entity after the verification is passed, generating an AI identity through a national secret algorithm or other encryption algorithm, and limiting the access range of the AI entity according to an AI behavior white list rule;

[0050] S5, monitoring API calling behavior of the AI entity in real time, judging whether there is an abnormality based on a deep learning algorithm, and blocking the access of the AI entity and triggering an alarm if there is an abnormality;

[0051] S6, processing business data, encrypting and transmitting the business data by using a national secret algorithm or other encryption algorithm, encrypting and storing the business data by using the national secret algorithm or other encryption algorithm, and verifying data integrity by using the national secret algorithm or other encryption algorithm.

[0052] Specifically, in practical applications, the server side can be a high-performance computing device deployed on an enterprise private cloud or public cloud, such as a server cluster supporting multi-core CPU and GPU acceleration, and its operating system can be Linux or Windows Server to meet the high-concurrency data processing requirements. In step S1, the user identity data sent by the client is transmitted in JSON format through the HTTPS protocol, and the biometric data (such as iris images) is uploaded after preprocessing (such as grayscale and feature point extraction), and the behavior data is recorded by the client log collection module to record the user's keyboard input frequency, mouse movement trajectory and other dynamic information. In step S2, the national encryption algorithm is used to generate ePASS-ID, which uses a hash algorithm that meets the national encryption standard to generate a unique identifier of fixed length, ensuring collision resistance; the national encryption algorithm also includes identity-based encryption technology, which assigns a key to each user through a key generation center (KGC), and the encrypted biometric data is stored in a distributed database (such as MongoDB) to improve query efficiency. The organizational structure information database in step S3 can be an enterprise employee information table built based on a relational database (such as MySQL), and the digital appointment file is stored in XML format, containing fields such as position, department, and permission level; the four-dimensional permission control model realizes dynamic allocation logic through a scripting language (such as Python), supporting real-time updating and revoking of permissions. In step S4, the AI entity registration request is submitted through a special API interface, and the request data includes the meta information of the AI entity (such as model type, deployment environment) and digital signature, which is generated using the RSA algorithm (2048-bit key). The server side verifies the legality of the digital signature through a pre-set trust root certificate, and only allows registration requests from trusted sources (such as enterprise internal AI services) to pass. If the verification fails, the system records the request IP and adds it to the temporary blacklist to prevent subsequent attempts. After verification, the server side requires the AI entity to perform a standardized test task in a controlled environment, which is a dynamic challenge sequence including 100 pre-defined API call requests (such as GET / api / data, POST / api / update, input is fixed JSON format {"query":"test"} ) and 15 adversarial challenge requests. The adversarial challenge request is generated based on the historical behavior template of the legal AI within the enterprise (extracted from the Elasticsearch database, covering 100,000 calls of 1,000 AI entities), which includes statistical features of API call sequences (such as response time mean, error rate distribution).The challenge request includes non-standard parameters (e.g. malformed JSON {"query": null}), high-load computing tasks (e.g. matrix operation API), and fake behavior simulation (e.g. abnormally high frequency of calls, 10 times per second). Legitimate AI can handle these tasks efficiently due to optimization for enterprise APIs such as fine-tuning or caching mechanisms, with an error rate of less than 5% and stable response time (mean < 200 milliseconds). Fake AI exhibits high error rates (> 20%) or inconsistent resource usage (memory fluctuations > 30%) due to lack of optimization. The test tasks run in isolated Docker containers (4-core CPU, 8 GB memory, Ubuntu 20.04), with behavior logs recorded for each call. The logs include response time (milliseconds), memory usage (MB), CPU usage (%), and API call error rate (proportion of error responses). The behavior analysis algorithm uses k-means clustering to extract model fingerprint data. K-means is chosen for its efficiency and robustness to high-dimensional data, allowing it to extract compact AI computing feature representations from behavior logs. The specific steps are as follows: 1) Normalize response time, memory usage, CPU usage, and error rate to [0, 1]; 2) Represent each log record as a four-dimensional feature vector to form a feature matrix of 115 records (100 API calls + 15 challenge requests); 3) Apply k-means algorithm (k = 5, based on elbow rule, k-means++ initialization, maximum iterations 100 times, convergence error threshold 0.001) to generate 5 cluster centers; 4) Concatenate the cluster centers into a 20-dimensional vector (5 x 4 dimensions) as fingerprint data. Fingerprint data reflects AI's computing characteristics (such as response time stability and error rate tendency). Legitimate AI's fingerprint vector forms a compact cluster (intra-class distance < 0.1) due to low error rate and high stability, while fake AI's fingerprint vector is an outlier (inter-class distance > 0.3) due to high error rate and resource anomalies. To verify the fingerprint differentiation ability, the system calculates the Euclidean distance between the fingerprint vector and the legitimate AI template, and the mean distance of fake AI exceeds 2 standard deviations. The test task is repeated 3 times, and the fingerprint mean is taken to ensure consistency. If there are less than 90 logs, the system rejects registration and records an exception. Fingerprint data is converted into a 32-byte AI identity using the national encryption algorithm (SHA-256) and stored in MongoDB. The whitelist rule is based on the function range (such as "read-only operation") and fingerprint data, in JSON format, with Redis cache supporting 100,000 queries per second. If registration fails, the system sends an alert via SMTP.Specifically, the real-time monitoring AI behavior sandbox is built based on Linux's cgroups and namespace technology, and each AI entity is allocated independent CPU and memory resources (2 cores and 4 GB) to prevent resource competition. A lightweight agent program (written in Go language) runs inside the sandbox, which collects API call logs in real time, including call time, interface name, parameters, and error rate (defined from step S4 challenge task), and logs are transmitted to the server side through Kafka and stored in Elasticsearch. The anomaly detection uses a long short-term memory network (LSTM) model trained in the PyTorch framework, with a binary classification (normal vs. abnormal) as the training target. The training samples include: 1) normal samples: 100,000 API calls of legitimate AI within the enterprise (30 days, 1000 AI entities), features include call time, interface name, parameter hash value, error rate (introduced from S4, <5%), and labeled "normal"; 2) abnormal samples: 5500 records generated by red team testing, covering high-frequency probes (1000), unauthorized access (1000), GAN forgery (1000), low-frequency forgery (500), and mixed attacks (1000, such as forgery + unauthorized access), with an error rate >20% and labeled "abnormal". The samples are stored in Elasticsearch. To adapt to LSTM, the data is preprocessed as follows: call time is normalized to [0, 1] relative time interval; interface name is converted to a 64-dimensional vector through Word2Vec (based on 1000 API interface names, window size 5, 64-dimensional vector, 50 epochs, and enterprise API documentation as corpus); parameter hash value (256-bit SHA-256) is reduced to 32 dimensions (retaining 90% variance) through PCA; error rate is directly used as a 1-dimensional feature; and the 97-dimensional feature vector (64+32+1) is formed, combined with a 50-call window, and mapped to a 128-dimensional input vector through a fully connected layer. The training uses cross-entropy loss, L = -[y·log(ŷ) + (1-y)·log(1-ŷ)], where y is the label (0 normal, 1 abnormal). The model is a three-layer LSTM (256 hidden units per layer, determined by grid search, balancing accuracy and efficiency), with a training set:validation set = 8:2, Adam optimizer (learning rate 0.001), 100 epochs, accuracy 95%, and F1 score 0.93. The 50-call window is optimized through experiments, balancing temporal information and computational efficiency. In real-time detection, the system extracts a 50-call sequence, converts it to a 128-dimensional vector according to the preprocessing, and outputs an anomaly probability (threshold 0.7) through LSTM. If it is abnormal, it is blocked through iptables, and a WebSocket alert is pushed (including the type of anomaly, timestamp). If the log is less than 50 times, fill in the zero vector or pause the detection. Secondary biometric verification (iris or voiceprint, 30-second timeout, permanent ban on failure) is triggered when an anomaly occurs.In step S6, the national cryptographic algorithm and other encryption algorithms are used for asymmetric encryption of data transmission, the key length conforms to the national standard, ensuring the security of transmission; the national cryptographic algorithm is also used for symmetric encryption, and the data is encrypted in blocks and stored in a local encrypted file system (such as LUKS); the national cryptographic algorithm checks data integrity, calculates a check value when data is written and read, and verifies consistency to prevent tampering; in step S5, the implementation of abnormality judgment based on deep learning algorithm relies on a pre-trained long short-term memory (LSTM) model, and the training samples of the model come from the following two types of data:

[0053] Normal samples: generated by historical API call logs of legitimate AI entities (such as intelligent customer service, data analysis models) within the enterprise, the logs include call time, interface name, parameter hash value, etc. Features, each record is labeled as "normal", the sample collection period is 30 days, covering 100,000 calls of at least 1,000 AI entities, and the data is stored in an Elasticsearch database.

[0054] Abnormal samples: generated by simulating attack scenarios, including API calls with fake identities (such as high-frequency probes, unauthorized access), fake behavior data (such as simulating user mouse tracks), and known malicious AI behavior patterns (such as call sequences based on open-source malicious scripts), abnormal samples are generated through red team testing, simulating attacks covering GAN forgery, SQL injection, etc. Each type of abnormal sample contains at least 5,000 records, labeled as "abnormal".

[0055] The training process uses the PyTorch framework, the model input is a vector representation of the API call sequence (dimension 128), the three-layer LSTM network extracts time series features, and the output is a binary classification result (normal / abnormal). The training set and validation set are divided according to 8:2, and the accuracy reaches more than 95% after 100 epochs of training. When judging abnormalities, the system extracts the API call sequence of the AI entity in real time, compares it with the trained model, and if the classification result is abnormal, the blocking and alarm mechanism is triggered.

[0056] The biometric data in step S1 includes face, iris, voiceprint, and finger vein data, the behavior data includes user operation habit data, and the user identity data is verified through a multi-modal credential system.

[0057] Specifically, the biometric data collection device can include an iris scanner (such as IrisGuard IG-AD100), a voiceprint input microphone (such as Shure SM58), and a finger vein recognition instrument (such as Hitachi H1), which are connected to the client through USB or Bluetooth, and the collected data formats are JPEG (iris), WAV (voiceprint), and binary stream (finger vein) respectively. The iris data extracts the iris boundary through Hough transformation, the voiceprint data extracts the feature through Mel Frequency Cepstral Coefficient (MFCC), and the finger vein data generates the blood vessel texture map through infrared imaging technology. The behavior data collection module runs on the client operating system (such as Windows 10 or Android), captures the keyboard typing interval and mouse click coordinates through the hook function (such as SetWindowsHookEx of Windows API), generates time series data, and the sampling frequency can be set to 10 times per second. The multi-modal credential system combines biometric and behavioral data for dual verification. First, the biometric data is preliminarily screened through the support vector machine (SVM) algorithm, and after successful matching, the consistency of the behavior data is analyzed using the hidden Markov model (HMM). If both pass, the identity is confirmed to be legal. The advantage of multi-modal verification is to improve the robustness of identity authentication. For example, when iris recognition fails due to insufficient light, voiceprint and behavior data can still be used as supplementary evidence.

[0058] The four-dimensional permission control model in step S3 includes:

[0059] An organizational structure dimension generates a permission inheritance relationship according to the enterprise hierarchical structure;

[0060] A data sensitivity dimension encrypts data based on a national encryption algorithm or other encryption algorithms;

[0061] A business scenario dimension encrypts different business environments using a national encryption algorithm or other encryption algorithms;

[0062] A time-space dimension restricts access permissions according to GPS and IP addresses.

[0063] Specifically, the implementation of the four-dimensional permission control model relies on a multi-level permission management system deployed on the server side, developed using the Java Spring framework, and supporting RESTful interface calls. The organizational structure dimension stores enterprise hierarchical information through a tree structure, such as department A subordinate team B. The permission inheritance relationship is automatically calculated through SQL query statements (such as WITH RECURSIVE), ensuring that child nodes inherit the permissions of parent nodes. The business scenario dimension creates independent national encryption algorithm encryption sandboxes through Docker container technology, with each sandbox assigned a unique key and IP address. Sandboxes are isolated through a virtual network, such as using Kubernetes' NetworkPolicy to restrict cross-scenario access. The data sensitivity dimension divides data into three levels: public, internal, and confidential. The hierarchical keys generated by the national encryption algorithm are stored in a hardware security module (HSM, such as YubiHSM). The encryption process supports batch processing, with 1000 records processed per second. The time-space dimension combines the client's GPS coordinates (accuracy ± 10 meters) and IP address (resolved through a GeoIP database), and the permission rules are stored in JSON format. Access requests outside the scope will be denied.

[0064] The AI behavior whitelist rules in step S4 include:

[0065] Registering the model fingerprint data of AI entities to limit their data interaction range;

[0066] Prohibiting AI entities from accessing sensitive data beyond their ePASS-ID corresponding permissions.

[0067] Specifically, the generation process of AI behavior whitelist rules includes two stages: AI entity registration and rule matching. During registration, AI entities need to submit model fingerprint data, such as deep learning models trained based on PyTorch can export parameters through torch.save. The server side uses national encryption algorithms to generate fingerprint hash values as AI identity identifiers. The data interaction range is limited through configuration files and stored in the AI permission collection of MongoDB. The matching of whitelist rules is implemented using the Trie tree algorithm, supporting fast prefix queries. For example, the rule for prohibiting access to sensitive data tables (such as user information tables) can be written as "deny:select:user_*", and 100,000 matching requests per second can be handled. The implementation of prohibiting access to sensitive data beyond permissions relies on the database's access control list (ACL), combined with AI identity identification for double verification. If unauthorized behavior is detected, the system will record logs and send alert emails to administrators through the SMTP protocol.

[0068] Real-time monitoring in step S5 includes:

[0069] Isolating AI entity operations in the AI behavior sandbox;

[0070] If an abnormal API call is detected, a secondary biometric verification is triggered.

[0071] Specifically, the real-time monitoring AI behavior sandbox is built based on Linux's cgroups and namespace technology, and each AI entity is allocated independent CPU and memory resources (such as 2 cores and 4 GB), preventing resource competition. A lightweight agent program (written in Go language) runs inside the sandbox, collecting API call logs in real time, including call time, interface name, and parameters. The logs are transmitted to the server-side analysis through the Kafka message queue. The deep learning model for anomaly detection is based on the PyTorch framework, and the training data set includes normal API call samples (such as GET / api / data) and abnormal samples (such as high-frequency POST / api / admin), with a model accuracy of over 95%. If an anomaly is detected, such as API call frequency exceeding 100 times per second, the system blocks the AI entity's network connection through iptables rules, while triggering a secondary biometric verification. The secondary verification pushes a notification to the client (based on the WebSocket protocol), requiring the user to resubmit iris or voice data, with a verification timeout of 30 seconds. If it fails, the AI entity is permanently banned from access.

[0072] Step S6 further includes:

[0073] Adding a national cryptographic algorithm or other encryption algorithm hash digital watermark to the exported data;

[0074] Setting up a multi-level approval process for sensitive data operations and storing evidence on the chain.

[0075] Specifically, the digital watermark generated by the national cryptographic algorithm is implemented by calculating the hash value of the exported data (such as a PDF document), and then embedding the hash value in the file metadata or invisible area, such as the XMP field of a PDF. The embedding process is completed through the open-source library PyMuPDF. During watermark verification, the system recalculates the file hash and compares it with the embedded value to ensure that the data has not been tampered with. The multi-level approval process is implemented based on a workflow engine (such as Activiti), for example, sensitive data export requires two levels of approval by the department manager and the security officer, and the approval records are stored on the chain after being summarized by a hash algorithm. The Hyperledger Fabric blockchain platform is used for the chain storage process, and the data is stored in JSON format, with each record occupying about 1 KB of storage space. The advantage of chain storage is to provide tamper-proof audit logs, suitable for scenarios with high compliance requirements, such as data management by government agencies. In addition, the multi-level approval process supports dynamic adjustment, such as automatically increasing the approval level based on data sensitivity, ensuring flexibility and security of the process.

[0076] To protect against AI entity identity forgery, the present embodiment works through the following closed-loop mechanism:

[0077] Identity pre-verification: In step S4, AI entities are registered and verified through digital signatures and behavior logs to ensure that only legitimate AI enters the system, excluding the initial risk of fake identities.

[0078] Behavioral feature monitoring: Step S5 uses deep learning models to analyze API call sequences and combines model fingerprint data to dynamically detect fake behaviors (such as simulating the call patterns of legitimate AI).

[0079] Quick response mechanism: After detecting abnormalities, the system blocks access through iptables and triggers secondary biometric verification to prevent fake AI from bypassing protection through temporary identities.

[0080] For example, in an enterprise deployment scenario, legitimate AI (such as an automatic report generation model) generates a unique identifier through registration, and its API call pattern complies with the whitelist rules. When an external malicious AI attempts to fake an identity, its call sequence is identified as abnormal by the deep learning model due to the lack of legitimate behavior patterns, and the system completes the block within 500 milliseconds, with alarm information sent to the administrator through SMTP.

[0081] Example Two:

[0082] Reference Figure 2 In the second embodiment of the present application, the present application provides a data security protection system based on anti-AI penetration technology, configured on the server side, comprising:

[0083] A receiving module for receiving user identity data and AI entity registration requests sent by the client;

[0084] An identity generation module for generating ePASS-ID based on user identity data and generating an identity for AI entities;

[0085] A permission allocation module for generating and allocating permission data based on organizational structure information database;

[0086] A monitoring module for real-time monitoring of API call behavior of AI entities and determining abnormalities;

[0087] A data processing module for encrypted transmission and storage of business data.

[0088] Specifically, the data security protection system deployed on the physical hardware of the server end can be a rack-mounted server supporting a RAID5 disk array, equipped with at least 64 GB of memory and 1 TB of SSD storage to meet the high concurrent access requirements. The receiving module realizes load balancing through Nginx reverse proxy, supports processing 5000 client requests per second, and transmits user identity data and AI registration requests in Protobuf format to reduce bandwidth occupation. The identity generation module runs in a separate microservice instance (based on SpringBoot), accelerates ePASS-ID generation through Redis cache, and can generate 10,000 identities per second; the AI identity generation is linked with the AI model fingerprint database (PostgreSQL) to ensure consistency. The permission allocation module synchronizes organizational structure information through a timing task (CronJob), updates permission data daily, and supports 100,000 users. The monitoring module integrates Prometheus and Grafana tools to display API call frequency and abnormal rate in real time, and the abnormal judgment logic is deployed as a separate Python script to support dynamic loading of new rules. The data processing module implements data encryption based on the national cryptographic algorithm and other encryption algorithms, with an encryption speed of up to 50 MB per second, and the storage process realizes high availability through a distributed file system (such as HDFS).

[0089] The identity generation module generates ePASS-ID through the national cryptographic algorithm or other encryption algorithms, and encrypts the biometric data using the national cryptographic algorithm or other encryption algorithms.

[0090] Specifically, the specific implementation of the identity generation module includes an encryption service sub-module running in a Docker container on the server end, and the container image is built based on Ubuntu 20.04 and pre-installed with the national cryptographic algorithm library. The input of the national cryptographic algorithm is the binary stream of the user identity data, such as the iris feature vector (about 2 KB), and the output is a fixed-length hash value, with a generation process time of about 5 milliseconds. The national cryptographic algorithm also includes identity-based encryption technology, and the key pair is generated by the KGC, with a key length conforming to national standards, and the private key is stored in the HSM to prevent leakage. The encrypted biometric data is saved in Base64 encoded format to the database, supporting fast retrieval, such as querying the corresponding record within 1 millisecond through the index field "user_id". The module supports multi-threaded processing, and can handle 1000 encryption requests per second, suitable for high-load scenarios. In addition, the identity generation module provides error handling mechanisms, such as returning a JSON format error code if the input data format is incorrect, ensuring system robustness.

[0091] Embodiment Three

[0092] The third embodiment of the present application is based on the same inventive concept, and the present application provides a computer readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the data security protection method based on the anti-AI penetration technology of the above-mentioned embodiments.

[0093] Embodiment four

[0094] The fourth embodiment of the present application is based on the same inventive concept, and the present application provides a computer device, which comprises a processor and a memory. The processor and the memory are in communication with each other. The memory is used to store instructions, and the processor is used to execute the instructions in the memory to implement the data security protection method based on the anti-AI penetration technology of the above-mentioned embodiments.

[0095] It should be understood that parts of the present application can be realized by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be realized by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if realized by hardware, and as in another embodiment, it can be realized by any one or a combination of the following technologies known in the art: discrete logic circuit with logic gate circuit for implementing logic function on data signal, special integrated circuit with suitable combination logic gate circuit, programmable gate array (PGA), field programmable gate array (FPGA) and the like.

[0096] Finally, it should be noted that: the above only describes the preferred embodiments of the present application and is not used to limit the present application, although the present application has been described in detail with reference to the foregoing embodiments, for those skilled in the art, the technical solutions recorded in the foregoing embodiments can still be modified, or some technical features can be replaced by equivalents, any modification, equivalent replacement, improvement and the like within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A data security protection method based on anti-AI penetration technology, characterized in that, Applied to the server side, comprising the following steps: S1, receiving the user identity data sent by the client, the user identity data including biometric data and behavior data; S2, based on the user identity data, generating a unique ePASS-ID through a national encryption algorithm or other encryption algorithm, and encrypting and storing the biometric data using a national encryption algorithm or other encryption algorithm; S3, according to the pre-set organizational structure information database, parsing the digital appointment file, generating the permission data corresponding to the ePASS-ID, and the permission data is dynamically allocated based on the four-dimensional permission control model; S4, receiving the registration request of the AI entity, pre-verifying the identity of the AI entity, obtaining its model fingerprint data after verification, generating an AI identity through a national encryption algorithm or other encryption algorithm, and limiting the access range of the AI entity according to the AI behavior white list rule; S5, real-time monitoring of the API calling behavior of the AI entity, judging whether there is an abnormality based on a deep learning algorithm, and if there is an abnormality, blocking the access of the AI entity and triggering an alarm; S6, processing business data, encrypting transmission using a national encryption algorithm or other encryption algorithm, encrypting storage using a national encryption algorithm or other encryption algorithm, and verifying data integrity through a national encryption algorithm or other encryption algorithm; The four-dimensional permission control model in step S3 includes: Organizational structure dimension, generating permission inheritance relationship according to enterprise hierarchical structure; Data sensitivity dimension, hierarchical encryption of data based on a national encryption algorithm or other encryption algorithm; Business scenario dimension, using a national encryption algorithm or other encryption algorithm to encrypt sandbox to isolate different business environments; Time-space dimension, limiting access permissions according to GPS and IP address.

2. The data security protection method based on anti-AI penetration technology according to claim 1, characterized in that, The biometric data in step S1 includes face, iris, voiceprint and finger vein data, the behavior data includes user operation habit data, and the user identity data is verified through a multi-modal certificate system.

3. The data security protection method based on anti-AI penetration technology according to claim 1, characterized in that, The AI behavior white list rule in step S4 includes: Register the model fingerprint data of the AI entity to limit its data interaction range; Prohibit the AI entity from accessing sensitive data beyond the permission corresponding to its ePASS-ID.

4. The data security protection method based on anti-AI penetration technology according to claim 1, characterized in that, The real-time monitoring in step S5 includes: Isolating the operation of the AI entity in the AI behavior sandbox; If an abnormal API call is detected, triggering a secondary biometric feature verification.

5. The data security protection method based on anti-AI penetration technology according to claim 1, characterized in that, The step S6 further includes: Adding a national encryption algorithm or other encryption algorithm hash digital watermark to the exported data; Setting up a multi-level approval process for sensitive data operations and chaining for storage.

6. A data security protection system based on anti-AI penetration technology, characterized in that, The data security protection method based on anti-AI penetration technology according to any one of claims 1-5 is configured on the server side, comprising: A receiving module for receiving user identity data sent by a client and a registration request of an AI entity; An identity generation module for generating an ePASS-ID based on the user identity data and generating an identity for an AI entity; A permission allocation module for generating and allocating permission data according to an organizational structure information database; A monitoring module for real-time monitoring of the API calling behavior of the AI entity and judging abnormalities; A data processing module is configured to encrypt and store the business data.

7. The data security system based on anti-AI penetration technology according to claim 6, characterized in that, The identity generation module generates the ePASS-ID by using a national encryption algorithm or other encryption algorithm, and encrypts the biometric data by using the national encryption algorithm or other encryption algorithm.

8. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the computer program to realize the data security protection method based on the anti-AI penetration technology according to any one of claims 1 to 5.

9. A readable storage medium, characterized by, The readable storage medium stores the computer program, and the computer program is executed by the processor to realize the data security protection method based on the anti-AI penetration technology according to any one of claims 1 to 5.

Images