An analyst observes an autonomous defensive system ingest an adversary-controlled artifact, message, or other lower-trust content during a security workflow. Following ingestion, the system unexpectedly terminates or truncates analysis, suppresses a detection, modifies a security decision, or attempts to disable a defensive capability contrary to the established workflow. The analyst correlates the ingested content, defensive task execution, and subsequent security-control state or disposition to identify potential adversarial manipulation of autonomous defenses.
| Field | Description |
|---|---|
| ExpectedDefensiveWorkflows | Required analysis, detection, triage, or investigation steps for each defensive task |
| ApprovedTerminationConditions | Authorized reasons for ending, truncating, or refusing analysis |
| RequiredAnalysisOutputs | Evidence, findings, and completion artifacts expected before a defensive task can be considered complete |
| AuthorizedAlertDispositions | Permitted alert state transitions and the conditions required for each |
| ProtectedSecurityControls | Defensive configurations, policies, and capabilities that the autonomous agent cannot disable or modify without authorization |
| ArtifactToDeviationWindow | Maximum interval between processing suspicious content and an anomalous defensive action |
| RequiredTaskCorrelation | Task, artifact, trace, and operation identifiers necessary to correlate the event sequence |
| ApprovedWorkflowExceptions | Legitimate maintenance, policy-driven exclusions, and authorized human interventions |