System and method for threat intelligence analysis based on multi-agent artificial intelligence

TWI946234BActive Publication Date: 2026-11-11CHUNGHWA TELECOM CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
TW114150121
Authority / Receiving Office
TW · TW
Patent Type
Patents
Current Assignee / Owner
Filing Date
2025-12-19
Publication Date
2026-11-11
Estimated Expiration
2045-12-18

Smart Images

  • Figure TWG2TB001920986_001
    Figure TWG2TB001920986_001
  • Figure TWG2TB001920986_002
    Figure TWG2TB001920986_002
  • Figure TWG2TB001920986_003
    Figure TWG2TB001920986_003
Patent Text Reader

Abstract

This invention proposes a threat intelligence analysis system and method based on multi-agent artificial intelligence. It establishes an automated decision-making process by dynamically assigning tasks and integrating agents for data collection, classification, research, hunting, evaluation, and report generation through an agent collaborative scheduling module. The system combines enhanced memory and semantic retrieval modules to ensure the standardization and traceability of cross-source data. It also integrates threat intelligence platforms, open-source intelligence, sandboxes, dark web, and enterprise cybersecurity systems by selecting large-scale language models locally or in the cloud through multi-model scheduling. By assessing confidence levels and risk values, it improves judgment accuracy, outputs differentiated reports, effectively enhances detection efficiency, and reduces labor costs.
Need to check novelty before this filing date? Find Prior Art

Claims

1. A threat intelligence analysis system based on multi-agent artificial intelligence, comprising: The agent collaboration unit includes: an agent collaboration scheduling module configured to read query prompts, wherein the query prompts are associated with a threat source; a threat collection agent connected to the agent collaboration scheduling module, configured to query multiple threat event data corresponding to the query prompts, and convert the multiple threat event data into multiple standardized data; a threat classification agent connected to the agent collaboration scheduling module, configured to determine threat indicators based on the multiple standardized data, and provide the threat indicators to the agent collaboration scheduling module; a threat research agent connected to the agent collaboration scheduling module, configured to access multiple malicious samples associated with the threat source, analyze the multiple malicious samples, and generate research results based on the analysis results of the multiple malicious samples; and a threat hunting agent connected to the agent collaboration scheduling module and the threat research agent, configured to query malicious activity records associated with the threat source in the enterprise information security system. An evaluation agent, connected to the agent coordination and scheduling module, is configured to update the threat indicators based on the research results and the malicious activity records, and to filter the research results and the malicious activity records to remove sensitive information. The evaluation agent is configured to perform checks on input and output data using Model Context Protocol hooks (MCP hooks). These checks include: removing sensitive personal data, performing content security checks, and performing risk detection on the model output, as well as filtering inappropriate content or unauthorized requests according to a preset content security policy. A report generation agent, connected to the agent coordination and scheduling module and the evaluation agent, is configured to generate a threat report based on the updated threat indicators and the filtered research results and malicious activity records.

2. The threat intelligence analysis system as described in claim 1, wherein each of the standardized data is in JSON format.

3. The threat intelligence analysis system as described in claim 1, wherein the threat indicators include at least one of confidence level indicators and risk indicators.

4. The threat intelligence analysis system as described in claim 1, wherein the threat research agent is configured to invoke multiple analysis tools within the analysis tool unit to analyze the multiple malicious samples.

5. The threat intelligence analysis system as described in claim 4, wherein the plurality of analysis tools include at least one of the following: malicious sample intelligence comparison tools, open source intelligence, malware sandbox analysis, vulnerability databases, binary program analysis tools, red team simulation tools, and industrial control / IoT security tools.

6. The threat intelligence analysis system as described in claim 1, wherein the enterprise security system includes at least one of a security information incident management system, a security coordination automated response system, a firewall / intrusion detection system, an extended detection and response system, and a cloud security posture management system.

7. The threat intelligence analysis system as described in claim 5 further includes an enhanced memory unit connected to the agent collaboration unit via a multi-model scheduling and management module. This unit is configured to perform data retrieval, preprocessing, semantic indexing, and vectorized storage, and to provide the threat gathering agent, the threat classification agent, the threat hunting agent, the assessment agent, and the report generation agent with contextual information associated with the threat source through semantic retrieval. This information serves as the basis for determining the threat indicators, studying the multiple malicious samples, or querying the malicious activity records.

8. The threat intelligence analysis system as described in claim 1 further includes a multi-model scheduling management module connected to the agent collaboration unit; wherein the multi-model scheduling management module is configured to assemble a context based on the query prompts, the threat indicators, or the feedback from the threat gathering agent, maintain a list of callable models and tools, and perform cross-model dynamic routing and security checks between local large language models and external large language models to support threat analysis and threat report generation.

9. The threat intelligence analysis system as described in claim 1 has a cyclical analysis mechanism; wherein, The assessment agent is configured to determine whether the information is sufficient or the risk is clear based on the threat indicators. If the threat indicators do not meet the preset judgment conditions, the agent coordination scheduling module is configured to automatically reassign the threat research agent and the threat hunting agent to perform additional analysis and verification to obtain updated research results or malicious activity records. Furthermore, the agent collaboration unit repeats the steps of updating the threat indicators and generating the threat report until the threat indicators meet the preset judgment conditions, forming a feedback decision loop.

10. The threat intelligence analysis system as described in claim 7, further comprising an interface for interaction between the threat intelligence analysis system and external systems, responsible for requesting access, authentication, traffic control and security checks, and converting the requirements of the external systems and the multiple analysis tools into an application interface gateway in an internal standard format, wherein the application interface gateway is used for external traffic and security boundary management, the multi-model scheduling management module is used for internal agent and model scheduling and context assembly, and the application interface gateway includes an authentication and authorization module, a traffic management and load balancing module, and a logging and monitoring module.

11. A threat intelligence analysis method based on multi-agent artificial intelligence, comprising: The agent-coordinated scheduling module reads query suggestions, wherein the query suggestions are associated with the threat source; The process involves several steps: a threat gathering agent queries multiple threat event data corresponding to the query prompts and converts these data into standardized data; a threat classification agent determines threat indicators based on the standardized data and provides these indicators to the agent coordination and scheduling module; a threat research agent accesses and analyzes multiple malicious samples associated with the threat source, generating research results based on the analysis; a threat hunting agent queries the enterprise cybersecurity system for malicious activity records associated with the threat source; an assessment agent updates the threat indicators based on the research results and malicious activity records, and filters these records to remove sensitive information. The assessment agent uses Model Context Protocol hooks (MCP hooks) to check input and output data, including removing sensitive personal data, performing content security checks, risk detection on model output, and filtering inappropriate content or unauthorized requests according to preset content security policies; and a report generation agent generates a threat report based on the updated threat indicators and the filtered research results and malicious activity records.

Citation Information

Patent Citations

  • Automated vulnerability and threat landscape analysis

    US20230283629A1

  • Systems and methods for intelligent cyber security threat detection and mitigation through an extensible automated investigations and threat mitigation platform

    US20240414194A1