A time-isolation-domain-based multi-agent runtime security isolation and conflict resolution method and system
By combining time isolation windows and unidirectional non-replayable execution graphs, runtime security issues in multi-agent systems are addressed. This achieves isolation and conflict resolution of abnormal agent behavior, ensures the independence and security of decision paths, prevents state pollution and replay attacks, and improves system security and reliability.
Patent Information
- Application Number
- CN202610829028.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-09
- Publication Date
- 2026-07-17
AI Technical Summary
Multi-agent systems face runtime security risks such as conflict propagation, spread of abnormal or malicious agent behavior, decision path contamination, and replay attacks, and lack effective security isolation and auditing mechanisms.
By employing time isolation window technology, the propagation of abnormal agent states is blocked, state changes are redirected to an independent snapshot space, and security adjudication and merging are achieved through a one-way non-replayable execution graph. Combined with conflict monitoring and independent inference snapshots, a global security audit mechanism is formed.
It effectively blocks the spread of abnormal agent states, prevents state pollution, resists replay attacks, improves system robustness, and ensures the credibility and security of decision results.
Smart Images
Figure CN122419974A_ABST