Multi-level dynamic isolation data processing system and method

Abstract

The invention discloses a multi-level dynamic isolation data processing system and method, and relates to the technical field of data security, the system comprises a data access module supporting a cloud native environment and a hybrid deployment mode, a strategy matching module, a sandbox execution module and an audit analysis module; the data access module receives the data, evaluates the sensitivity level and the purpose of the data, and labels the data based on an evaluation result; the strategy matching module reads the data label, searches an isolation strategy and an auditing rule which are matched with the data label from a strategy library, and automatically loads the strategies to the sandbox execution module once the matched strategies are found; the sandbox execution module selects an isolation mode according to the sensitivity level of the data, and processes and analyzes the data in an isolation environment according to a preset rule and algorithm; and the auditing analysis module records all operations on the data in real time, and analyzes the operation compliance based on a preset auditing rule. The problems of large resource consumption, insufficient strategy flexibility, coarse behavior audit granularity and the like in the existing data sandbox technology can be solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data security, in particular to a multi-level dynamic isolation data processing system and method. BACKGROUND

[0002] Existing data sandbox technologies mostly adopt static isolation strategies (such as virtual machine or containerization technology), which have the problems of large resource consumption and insufficient strategy flexibility. Static isolation technology needs to allocate fixed computing resources for each sandbox instance, which is prone to cause resource idling or overload when the business load fluctuates, and the isolation strategy needs to be manually predefined, which is difficult to dynamically adjust resource allocation and isolation rules according to real-time business scenarios.

[0003] The data usage behavior audit granularity is coarse, and it is difficult to achieve fine-grained abnormal behavior detection. Existing audit mechanisms mostly record user-level operation logs (such as data query, export), lack of fine tracking of operation parameters (such as query conditions, number of returned results), data flow (such as data transfer between components in the sandbox), and cannot identify hidden abnormal behaviors such as "high-frequency small amount of data theft".

[0004] Existing solutions lack support for dynamically changing business needs (such as temporary permission adjustment, multi-tenant isolation). In the multi-tenant scenario, a fixed role permission allocation mode is adopted, which cannot dynamically generate temporary permission groups according to the project cycle, and the resource isolation strategy of different tenant sandboxes is difficult to adjust in real time, which is prone to cause isolation boundary failure or resource contention problems when the tenant business scale rapidly expands. SUMMARY

[0005] The present application provides a multi-level dynamic isolation data processing system and method to solve the problems of large resource consumption, insufficient strategy flexibility and coarse behavior audit granularity in existing data sandbox technologies.

[0006] In the first aspect, the present application provides a multi-level dynamic isolation data processing system, which solves the above technical problems by adopting the following technical solutions:

[0007] A multi-level secure data sandbox system based on dynamic strategy, which includes a data access module supporting cloud-native environment and hybrid deployment mode, a strategy matching module, a sandbox execution module, and an audit analysis module, wherein:

[0008] The data access module receives data and evaluates the data sensitivity level and purpose, and labels the data based on the evaluation results;

[0009] The policy matching module comprises a dynamic policy engine, which realizes dynamic matching and loading of data tags and policies based on policy-as-code architecture through policy rules written in a declarative language managed by Git and stored in a code repository; in the matching process, the dynamic policy engine optimizes the matching policy through risk scores generated by a machine learning model in combination with user behavior logs, context environment data and historical risk events collected in real time by the data access module, and finally forms comprehensive decision rules including static policies and dynamic temporary policies and loads them into the sandbox execution module;

[0010] The sandbox execution module selects an isolation mode according to the sensitivity level of the data, and processes and analyzes the data in the isolated environment according to the preset rules and algorithms;

[0011] The audit analysis module records all operations on the data in real time, analyzes the compliance of the operations based on preset audit rules, and automatically generates an audit report containing risk assessment to assist in ensuring the compliance and security of data operations.

[0012] Optionally, the dynamic policy engine involved is the core component of the policy matching module, and based on the policy-as-code architecture, policy rules are written in a declarative language, the rule structure includes tag matching conditions, policy actions and execution priorities, the rules are stored in the code repository in the form of code, and the version management, audit tracking and rollback operation of the policy are realized by means of Git;

[0013] The dynamic policy engine collects user behavior logs, context environment data and historical risk event records output by the data access module in real time, and predicts the risk score of the current access request through a pre-trained machine learning model;

[0014] Based on the risk score result, the dynamic policy engine automatically generates a temporary policy and integrates it with the preset static policy to form a comprehensive access decision rule; the decision logic is deployed in the microservice gateway or API layer in the Sidecar mode through a lightweight engine, and the distributed execution of the policy is realized based on the service mesh to ensure that the policy update takes effect immediately;

[0015] The dynamic policy engine records all access decision results and risk event data throughout the process, automatically triggers priority adjustment or parameter optimization of the policy rule through analysis of misjudgment / omission cases, and inputs the labeled decision data into the machine learning model for incremental training to improve the accuracy of risk identification and the effectiveness of policy control, forming a positive cycle of "data-driven-model optimization-policy upgrade".

[0016] Optionally, the data access module involved specifically performs the following operations in sequence through the gateway service:

[0017] Receive data input from different data sources;

[0018] Based on the content, source or context of the input data, the sensitivity level and use classification of the data are evaluated;

[0019] According to the results of the sensitivity level and use classification, each piece of data is labeled accordingly.

[0020] Further optionally, the gateway service involved evaluates the sensitivity level based on the content, source or context of the input data, and divides the data into three levels of low sensitivity, medium sensitivity and high sensitivity through rule matching, quantitative scoring or machine learning models;

[0021] The sandbox execution module selects an isolation mode according to the sensitivity level of the data, and the isolation mode includes process level, container level and hardware virtualization, wherein: low sensitivity data is isolated at the process level, medium sensitivity data is isolated at the container level, and high sensitivity data is isolated at the hardware virtualization level.

[0022] Preferably, in the process level isolation environment, the low sensitivity data is processed according to the preset rules, the public analysis algorithm is directly called to generate a visual report, and the processing result is directly output to the non-sensitive data display layer;

[0023] In the container level isolation environment, the customized processing flow is started for the medium sensitivity data: first, part of the sensitive fields are processed through the data desensitization algorithm, then the transmission data is encrypted through the encryption engine, and finally the complex analysis is executed through the distributed computing framework, and the processing result is output to the internal business system after the compliance verification;

[0024] In the hardware virtualization isolation environment, the high sensitivity data is implemented with full-process security control: first, the data is loaded through the trusted execution environment to ensure that the data loading process is not tampered with; second, the federated learning or homomorphic encryption privacy computing technology is used to complete the model training without leaking the original data; the data flow is monitored in real time during the processing, and the processing result is stored in the special encryption storage cluster after multiple encryption, and only authorized users are allowed to access through the security gateway.

[0025] In the second aspect, the present application provides a multi-level dynamic isolation data processing method, which solves the above technical problems by adopting the following technical solutions:

[0026] A multi-level dynamic isolation data processing method, which supports cloud native environment and hybrid deployment mode, includes data access stage, policy matching stage, sandbox running stage and audit analysis stage, each stage performs the following operations:

[0027] S1, the data access stage: receiving data and evaluating the sensitivity level and use of the data, and labeling the data based on the evaluation results;

[0028] S2, policy matching stage: the dynamic policy engine is used as the core component of this stage, and the dynamic policy engine realizes the dynamic matching and loading of data tags and policies based on policy code architecture and policy rules written by declarative language managed by Git and stored in the code repository; in the matching process, the dynamic policy engine optimizes the matching policy by combining the user behavior logs, context environment data and historical risk events collected in real time in the data access stage, and the risk score generated by the machine learning model, and finally forms a comprehensive decision rule including static policy and dynamic temporary policy and loads it into the sandbox running stage;

[0029] S3, sandbox running stage: according to the sensitivity level of the data, the isolation mode is selected, and the data is processed and analyzed in the isolated environment according to the preset rules and algorithms;

[0030] S4, audit analysis stage: all operations on data are recorded in real time, the operation compliance is analyzed based on the preset audit rules, and the audit report including risk assessment is automatically generated to assist in ensuring the compliance and security of data operation.

[0031] Optionally, the dynamic policy engine involved is based on policy code architecture, and policy rules are written in declarative language, the rule structure includes tag matching condition, policy action and execution priority, the rules are stored in the code repository in the form of code, and the version management, audit tracking and rollback operation of the policy are realized by Git;

[0032] The dynamic policy engine collects the user behavior logs, context environment data and historical risk event records output by the data access module in real time, and predicts the risk score of the current access request through the pre-trained machine learning model;

[0033] Based on the risk score result, the dynamic policy engine automatically generates temporary policy and integrates with the preset static policy to form a comprehensive access decision rule; the decision logic is deployed in the micro-service gateway or API layer in the Sidecar mode through the lightweight engine, and the distributed execution of the policy is realized based on the service mesh, ensuring that the policy update takes effect immediately;

[0034] The dynamic policy engine records all access decision results and risk event data throughout the process, automatically triggers the priority adjustment or parameter optimization of the policy rule by analyzing the misjudgment / omission cases, and inputs the labeled decision data into the machine learning model for incremental training, improves the accuracy of risk identification and the effectiveness of policy control, and forms a positive cycle of "data-driven-model optimization-policy upgrade".

[0035] Optionally, the data access stage specifically executes the following operations through the gateway service in sequence:

[0036] S1.1, receiving data input of different data sources;

[0037] S1.2, based on the content, source or context of the input data, evaluating the sensitive level and use classification of the data;

[0038] S1.3, according to the results of the sensitive level and use classification, labeling each piece of data with the corresponding label.

[0039] Further, the gateway service involved evaluates the sensitive level based on the content, source or context of the input data, and divides the data into three levels of low sensitivity, medium sensitivity and high sensitivity through rule matching, quantitative scoring or machine learning model.

[0040] In the sandbox running stage, the isolation mode is selected according to the sensitive level of the data, and the isolation mode includes process level, container level and hardware virtualization, wherein: low sensitive data is isolated at process level, medium sensitive data is isolated at container level, and high sensitive data is isolated at hardware virtualization.

[0041] Preferably, in the process level isolation environment, the low sensitive data is processed according to the preset rules, the public analysis algorithm is directly called to generate the visual report, and the processing result is directly output to the non-sensitive data display layer.

[0042] In the container level isolation environment, the customized processing flow is started for the medium sensitive data: first, part of the sensitive fields are processed through the data desensitization algorithm, then the transmission data is encrypted by the encryption engine, and finally the complex analysis is executed through the distributed computing framework, and the processing result is output to the internal business system after the compliance verification.

[0043] In the hardware virtualization isolation environment, the high sensitive data is implemented with full-process security control: first, the data is loaded through the trusted execution environment to ensure that the data loading process is not tampered with; second, the federated learning or homomorphic encryption privacy computing technology is used to complete the model training without leaking the original data; the data flow is monitored in real time during the processing process, and the processing result is stored in the special encryption storage cluster after multiple encryption, and only authorized users are allowed to access through the security gateway.

[0044] The multi-level dynamic isolation data processing system and method of the present application has the following beneficial effects compared with the prior art:

[0045] The present invention uses multi-level dynamic isolation technology to select appropriate isolation methods according to the sensitivity level of the data, reduce unnecessary resource consumption, and improve the overall performance of the system; the dynamic policy engine adopts a policy-as-code architecture to realize declarative policy writing and version management, support real-time update and rollback of policies, and meet dynamically changing business needs; all operations on data are recorded in real time, the compliance of operations is analyzed based on preset audit rules, and audit reports containing risk assessments are automatically generated to assist in ensuring the compliance and security of data operations; it can solve the problems of high resource consumption, insufficient policy flexibility, and coarse granularity of behavioral auditing in existing data sandbox technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Attachment Figure 1 This is a module connection block diagram of the first embodiment of the present invention;

[0047] Attachment Figure 2 This is a flow chart of the method of embodiment 2 of the present invention. DETAILED DESCRIPTION

[0048] In order to make the technical solution, the technical problems solved and the technical effects of the present invention more clear, the technical solution of the present invention is clearly and completely described below in conjunction with specific embodiments.

[0049] Example 1:

[0050] Combined with attachment Figure 1 ,This embodiment proposes a multi-level security data sandbox system based on dynamic ,policies, which includes a data access module, a policy matching module, a sandbox ,execution module, and an audit analysis module that supports cloud native ,environment and hybrid deployment modes.

[0051] The data access module performs the following operations in sequence through the gateway service: receiving data input from different data sources; evaluating the sensitivity level and usage classification of the data based on the content, source or context of the input data; and labeling each piece of data accordingly based on the results of the sensitivity level and usage classification.

[0052] Specifically, the gateway service evaluates the sensitivity level based on the content, source, or context of the input data, and divides the data into three levels: low sensitivity, medium sensitivity, and high sensitivity through rule matching, quantitative scoring, or machine learning models.

[0053] The policy matching module comprises a dynamic policy engine, which realizes dynamic matching and loading of data tags and policies based on policy-as-code architecture through policy rules written in a declarative language managed by Git and stored in a code repository; in the matching process, the dynamic policy engine optimizes the matching policy through risk scores generated by a machine learning model in combination with user behavior logs, context environment data and historical risk events collected in real time by the data access module, and finally forms comprehensive decision rules comprising static policies and dynamic temporary policies and loads them into the sandbox execution module.

[0054] Specifically, the dynamic policy engine, as the core component of the policy matching module, writes policy rules in a declarative language based on policy-as-code architecture, the rule structure comprises tag matching conditions, policy actions and execution priorities, and the rules are stored in the code repository in the form of code, and the version management, audit tracking and rollback operation of the policy are realized with the help of Git;

[0055] The dynamic policy engine collects user behavior logs, context environment data and historical risk event records output by the data access module in real time, and predicts the risk score of the current access request through a pre-trained machine learning model;

[0056] Based on the risk score result, the dynamic policy engine automatically generates temporary policies and integrates them with preset static policies to form comprehensive access decision rules; the decision logic is deployed in the microservice gateway or API layer in a Sidecar mode through a lightweight engine, and the distributed execution of the policy is realized based on the service mesh to ensure that the policy update takes effect immediately;

[0057] The dynamic policy engine records all access decision results and risk event data throughout the process, automatically triggers priority adjustment or parameter optimization of the policy rules through analysis of misjudgment / omission cases, and inputs the labeled decision data into the machine learning model for incremental training to improve the accuracy of risk identification and the effectiveness of policy control, forming a positive cycle of "data-driven-model optimization-policy upgrade".

[0058] The sandbox execution module selects an isolation mode according to the sensitivity level of the data, and processes and analyzes the data in the isolated environment according to the preset rules and algorithms.

[0059] Specifically, the sandbox execution module selects an isolation mode according to the sensitivity level of the data, and processes and analyzes the data in the isolated environment according to the preset rules and algorithms.

[0060] In the process-level isolation environment, lightweight processing is performed on low-sensitive data according to preset rules, and a public analysis algorithm is directly called to generate a visual report, and the processing result is directly output to the non-sensitive data display layer.

[0061] In the container-level isolation environment, a customized processing flow is started for medium-sensitive data: first, some sensitive fields are processed through a data desensitization algorithm, then the transmission data is encrypted using a TLS encryption engine, and finally a complex analysis is performed through a distributed computing framework, and the processing result is output to the internal business system after compliance verification.

[0062] In the hardware virtualization isolation environment, full-process security management is implemented for high-sensitive data: first, the data is loaded through a trusted execution environment to ensure that the data loading process is not tampered with; second, privacy computing technologies such as federated learning or homomorphic encryption are used to complete model training without revealing the original data; real-time monitoring of data flow is performed during processing, and the processing result is stored in a special encrypted storage cluster after multiple encryption, and only authorized users are allowed to access it through a secure gateway.

[0063] The audit analysis module records all operations on data in real time, analyzes the compliance of the operations based on preset audit rules, and automatically generates an audit report containing risk assessment to assist in ensuring the compliance and security of data operations.

[0064] One thing to be added is that when data is transmitted between modules, a lightweight encryption channel is established, efficient encryption algorithms and key management mechanisms are used to ensure the security and integrity of data during transmission, and to prevent data from being intercepted or tampered with. Specifically, the lightweight encryption channel uses a pre-computation window or Montgomery ladder algorithm to reduce encryption delay, uses ECC to encrypt AES-GCM session keys, balances efficiency and security, implements low-delay encryption, and supports dynamic rotation and sharded storage of keys.

[0065] Embodiment Two:

[0066] In combination with the accompanying Figure 2 , the embodiment proposes a multi-level dynamic isolation data processing method, which supports cloud-native environment and hybrid deployment mode, including data access stage, policy matching stage, sandbox running stage and audit analysis stage, each stage specifically performs the following operations:

[0067] S1, the data access stage, through the gateway service, sequentially performs the following operations:

[0068] S1.1, receiving data input from different data sources;

[0069] S1.2, based on the content, source or context of the input data, evaluate the sensitivity level and usage classification of the data; in specific implementation, through rule matching, quantitative scoring or machine learning model, the data is divided into three levels of low sensitivity, medium sensitivity and high sensitivity;

[0070] S1.3, according to the results of sensitivity level and usage classification, label each piece of data accordingly.

[0071] S2, policy matching stage: the dynamic policy engine is used as the core component of this stage, and the dynamic policy engine realizes the dynamic matching and loading of data labels and policies based on the policy code architecture and the policy rules written by the declarative language managed by Git and stored in the code repository; In the matching process, the dynamic policy engine combines the user behavior logs, context environment data and historical risk events collected in real time in the data access stage, and optimizes the matching policy through the risk score generated by the machine learning model, and finally forms a comprehensive decision rule including static policy and dynamic temporary policy and loads it into the sandbox running stage.

[0072] Specifically, the dynamic policy engine is based on the policy code architecture, and the policy rules are written in declarative language. The rule structure includes label matching conditions, policy actions and execution priority. The rules are stored in the code repository in the form of code, and the version management, audit tracking and rollback operation of the policy are realized by Git;

[0073] The dynamic policy engine collects the user behavior logs, context environment data and historical risk event records output by the data access module in real time, and predicts the risk score of the current access request through the pre-trained machine learning model;

[0074] Based on the risk score result, the dynamic policy engine automatically generates temporary policy and integrates with the preset static policy to form a comprehensive access decision rule; the decision logic is deployed in the microservice gateway or API layer in the Sidecar mode through the lightweight engine, and the distributed execution of the policy is realized based on the service mesh, ensuring that the policy update takes effect immediately;

[0075] The dynamic policy engine records all access decision results and risk event data throughout the process, automatically triggers the priority adjustment or parameter optimization of the policy rule through the analysis of misjudgment / omission cases, and inputs the labeled decision data into the machine learning model for incremental training, improves the accuracy of risk identification and the effectiveness of policy control, and forms a positive cycle of "data driven-model optimization-policy upgrade".

[0076] S3, sandbox running stage: according to the sensitivity level of the data, select the isolation mode, and process and analyze the data in the isolated environment according to the preset rules and algorithms.

[0077] Specifically, the isolation modes include process level, container level and hardware virtualization, wherein: low-sensitive data is processed by process level isolation, medium-sensitive data is processed by container level isolation, and high-sensitive data is processed by hardware virtualization isolation.

[0078] In the process level isolation environment, the low-sensitive data is processed according to the preset rules, the public analysis algorithm is directly called to generate a visual report, and the processing result is directly output to the non-sensitive data display layer.

[0079] In the container level isolation environment, the medium-sensitive data starts a customized processing flow: first, part of the sensitive fields are processed by the data desensitization algorithm, then the transmission data is encrypted by the encryption engine, and finally the complex analysis is executed by the distributed computing framework, and the processing result is output to the internal business system after the compliance verification.

[0080] In the hardware virtualization isolation environment, the high-sensitive data is implemented with full-process security control: first, the data is loaded through the trusted execution environment to ensure that the data loading process is not tampered with; second, the federated learning or homomorphic encryption privacy computing technology is used to complete the model training without leaking the original data; the data flow is monitored in real time during the processing, and the processing result is stored in the special encryption storage cluster after multiple encryption, and only authorized users are allowed to access through the security gateway.

[0081] S4, audit analysis stage: all operations on the data are recorded in real time, the operation compliance is analyzed based on the preset audit rules, and the audit report containing risk assessment is automatically generated to assist in ensuring the compliance and security of data operation.

[0082] It should be noted that when the data is transmitted in each stage, a lightweight encryption channel is established, an efficient encryption algorithm and key management mechanism are used to ensure the security and integrity of the data during transmission, and to prevent the data from being intercepted or tampered with. Specifically, the lightweight encryption channel uses a pre-computation window or Montgomery ladder algorithm to reduce encryption delay, uses ECC to encrypt AES-GCM session keys, balances efficiency and security, realizes low-delay encryption, and supports dynamic rotation and sharded storage of keys.

[0083] As can be seen from the above, the multi-level dynamic isolation data processing system and method can solve the problems of large resource consumption, insufficient policy flexibility and coarse behavior audit granularity in existing data sandbox technology.

[0084] The principles and implementation manners of the present application are described in detail by using the above specific examples, and the examples are only used to help understand the core technical content of the present application. Based on the above specific examples of the present application, any improvement and modification of the present application made by the person skilled in the art without departing from the principles of the present application shall fall within the patent protection scope of the present application.

Claims

1. A multi-level dynamic isolation data processing system, characterized in that: It includes a data access module, a policy matching module, a sandbox execution module, and an audit analysis module that support cloud-native environments and hybrid deployment modes. The data access module receives data, assesses its sensitivity and usage, and labels the data based on the assessment results; The policy matching module includes a dynamic policy engine. Based on a policy-as-code architecture, the dynamic policy engine dynamically matches and loads data tags and policies using policy rules written in a declarative language that is versioned by Git and stored in a code repository. During the matching process, the dynamic policy engine combines user behavior logs, contextual data, and historical risk events collected in real time by the data access module, and optimizes the matching strategy using risk scores generated by machine learning models. Ultimately, it forms a comprehensive decision rule that includes both static and dynamic temporary policies and loads it into the sandbox execution module. The sandbox execution module selects an isolation method based on the sensitivity level of the data and processes and analyzes the data in an isolated environment according to preset rules and algorithms; The audit analysis module records all operations on data in real time, analyzes the compliance of operations based on preset audit rules, and automatically generates audit reports including risk assessments to help ensure the compliance and security of data operations.

2. A multi-level dynamically isolated data processing system according to claim 1, characterized in that: The dynamic policy engine, as the core component of the policy matching module, is based on a policy-as-code architecture and uses a declarative language to write policy rules. The rule structure includes tag matching conditions, policy actions, and execution priorities. The rules are stored in the code repository in code form, and Git is used to implement policy version management, audit tracking, and rollback operations. The dynamic policy engine collects user behavior logs, contextual environment data, and historical risk event records output by the data access module in real time, and uses a pre-trained machine learning model to predict the risk score of the current access request; Based on the risk scoring results, the dynamic policy engine automatically generates temporary policies and integrates them with the preset static policies to form comprehensive access decision rules. The decision logic is deployed in the microservice gateway or API layer in sidecar mode through a lightweight engine, and distributed execution of policies is achieved based on the service grid, ensuring that policy updates take effect immediately. The dynamic policy engine records all access decision results and risk event data throughout the process. By analyzing cases of misjudgment / omission, it automatically triggers priority adjustment or parameter optimization of policy rules, and inputs the labeled decision data into the machine learning model for incremental training, improving the accuracy of risk identification and the effectiveness of policy management, forming a positive cycle of "data-driven-model optimization-policy upgrade".

3. A multi-level dynamically isolated data processing system according to claim 1, characterized in that: The data access module specifically performs the following operations in sequence through the gateway service: Receive data input from different data sources; Assess the sensitivity level and usage classification of input data based on its content, source, or context; According to the results of sensitivity level and purpose classification, each piece of data is labeled accordingly.

4. A multi-level dynamic isolation data processing system according to claim 3, characterized in that: The gateway service assesses the sensitivity level of input data based on its content, source, or context, and classifies the data into three levels: low sensitivity, medium sensitivity, and high sensitivity through rule matching, quantitative scoring, or machine learning models; The sandbox execution module selects the isolation method according to the sensitivity level of the data. The isolation methods include process level, container level, and hardware virtualization. Among them, process level isolation is used for low-sensitivity data, container level isolation is used for medium-sensitivity data, and hardware virtualization isolation is used for high-sensitivity data.

5. A multi-level dynamic isolation data processing system according to claim 4, characterized in that: In a process-level isolation environment, lightweight processing is performed on low-sensitivity data according to preset rules, public analysis algorithms are directly called to generate visual reports, and the processing results are directly output to the non-sensitive data display layer; In a container-level isolated environment, a customized processing flow is initiated for moderately sensitive data: some sensitive fields are processed using a data desensitization algorithm, the transmitted data is then encrypted using TLS using an encryption engine, and complex analysis is performed using a distributed computing framework. The processing results are then output to the internal business system after compliance verification. In a hardware virtualization isolation environment, full-process security management and control is implemented for highly sensitive data: first, data is loaded through a trusted execution environment to ensure that the data loading process is not tampered with; second, privacy computing technologies such as federated learning or homomorphic encryption are used to complete model training without leaking the original data; data flow is monitored in real time during the processing process, and the processing results are stored in a dedicated encrypted storage cluster after multiple encryptions, and only authorized users are allowed to access through a secure gateway.

6. A data processing method for multi-level dynamic isolation, characterized in that: This method supports cloud-native environments and hybrid deployment modes, including data access, policy matching, sandbox operation, and audit analysis. Each stage performs the following operations: S1, Data access phase: Receive data and assess its sensitivity and usage, then label the data based on the assessment results; S2, Strategy Matching: The dynamic policy engine is the core component of this stage. Based on the policy-as-code architecture, the dynamic policy engine dynamically matches and loads data tags and policies through policy rules written in a declarative language that is versioned by Git and stored in a code repository. During the matching process, the dynamic strategy engine combines user behavior logs, contextual data, and historical risk events collected in real time during the data access phase, and optimizes the matching strategy using risk scores generated by machine learning models. Ultimately, it forms a comprehensive decision rule that includes both static strategies and dynamic temporary strategies and loads it into the sandbox runtime phase. S3, Sandbox operation stage: Select an isolation method based on the sensitivity level of the data, and process and analyze the data in the isolated environment according to preset rules and algorithms; S4, Audit and Analysis Phase: Record all operations on data in real time, analyze the compliance of operations based on preset audit rules, and automatically generate an audit report containing risk assessment to help ensure the compliance and security of data operations.

7. The data processing method of multi-level dynamic isolation according to claim 6, characterized in that: The dynamic policy engine is based on a policy-as-code architecture and uses a declarative language to write policy rules. The rule structure includes tag matching conditions, policy actions, and execution priorities. The rules are stored in the code repository in code form, and Git is used to implement policy version management, audit tracking, and rollback operations. The dynamic policy engine collects user behavior logs, contextual environment data, and historical risk event records output by the data access module in real time, and uses a pre-trained machine learning model to predict the risk score of the current access request; Based on the risk scoring results, the dynamic policy engine automatically generates temporary policies and integrates them with the preset static policies to form comprehensive access decision rules. The decision logic is deployed in the microservice gateway or API layer in sidecar mode through a lightweight engine, and distributed execution of policies is achieved based on the service grid, ensuring that policy updates take effect immediately. The dynamic policy engine records all access decision results and risk event data throughout the process. By analyzing cases of misjudgment / omission, it automatically triggers priority adjustment or parameter optimization of policy rules, and inputs the labeled decision data into the machine learning model for incremental training, improving the accuracy of risk identification and the effectiveness of policy management, forming a positive cycle of "data-driven-model optimization-policy upgrade".

8. The data processing method of multi-level dynamic isolation according to claim 6 is characterized in that: During the data access phase, the following operations are performed in sequence through the gateway service: S1.

1. Receive data input from different data sources; S1.

2. Evaluate the sensitivity level and usage classification of the input data based on its content, source, or context; S1.

3. Assign corresponding labels to each piece of data based on the sensitivity level and purpose classification results.

9. The data processing method of multi-level dynamic isolation according to claim 8, characterized in that: The gateway service assesses the sensitivity level of input data based on its content, source, or context. It categorizes the data into three levels: low, medium, and high sensitivity through rule matching, quantitative scoring, or machine learning models. During the sandbox operation phase, the isolation method is selected according to the sensitivity level of the data. The isolation methods include process level, container level, and hardware virtualization. Among them, process level isolation is used for low-sensitivity data, container level isolation is used for medium-sensitivity data, and hardware virtualization isolation is used for high-sensitivity data.

10. The data processing method of multi-level dynamic isolation according to claim 9, characterized in that: In a process-level isolation environment, lightweight processing is performed on low-sensitivity data according to preset rules, public analysis algorithms are directly called to generate visual reports, and the processing results are directly output to the non-sensitive data display layer; In a container-level isolated environment, a customized processing flow is initiated for moderately sensitive data: some sensitive fields are processed using a data desensitization algorithm, the transmitted data is then encrypted using TLS using an encryption engine, and complex analysis is performed using a distributed computing framework. The processing results are then output to the internal business system after compliance verification. In a hardware virtualization isolation environment, full-process security management and control is implemented for highly sensitive data: first, data is loaded through a trusted execution environment to ensure that the data loading process is not tampered with; second, privacy computing technologies such as federated learning or homomorphic encryption are used to complete model training without leaking the original data; data flow is monitored in real time during the processing process, and the processing results are stored in a dedicated encrypted storage cluster after multiple encryptions, and only authorized users are allowed to access through a secure gateway.

Images