AI Agent Goal Hijacking: Signals and Read-Only Goal Checks
Goal events concatenate the reported goal, plan, and decision.
AI agent containment
These guides document actual 1-SEC implementation paths and source fixtures. Start by verifying the event producer and field meanings, then read the coverage limits before choosing an enforcement policy.
AI Agent Tool Policy: Blocked Tools and Sensitive Targets
The built-in policy names blocked tools such as shell_exec, raw_sql, file_delete, and credential_access.
AI Agent Memory Poisoning and Cross-Session Signals
The memory monitor checks submitted writes for instruction-like content, a high write count within its window, and persistent/long_term memory associated with a changed session label.
Excessive Agency Monitoring: Scope, Approvals, Tools, and Plans
The LLM Firewall examines reported decisions/plans for actions without required approval, selected action keywords under readonly/read_only/limited scopes, large tool counts, complex plans, and elevated cross-agent delegation.
AI Agent Network Egress: Web-Fetch Signals and Real Network Controls
Web-fetch tracking inspects reported URLs/domains for sensitive destinations, unusual fetch patterns, external-to-internal transitions, and some suspicious URL parameter content.
Agent Rapid Actions and Rogue Loops: Alert Thresholds, Not Hard Budgets
The action tracker keeps per-agent action counts and repeated action/target state.
AI Agent Spawn Depth: Reported Parent Chains and Monitoring
Spawn handling computes a depth from reported agent/parent history, emits a new-agent observation, and increases severity when the computed depth exceeds three.
Cascading Agent Failures and Retry Storm Monitoring
Failure tracking records errors within a five-minute window and counts retry-like error types.
AI Agent Approval Bypass Events and Trust Boundaries
The trust-exploitation handler raises an impersonation alert from an impersonation event.
Shadow AI Monitoring: Known Endpoints and Reported Authorization
The known-endpoint check compares reported destination hosts with its implemented AI-service host list.
AI Agent Scope Drift: Sensitive-Target History and Its Limits
The action tracker records target strings per agent.