Skip to content

Adversarial AI & Autonomous Agent Compromise (AML.TA0001) — Detection Coverage Assessment

Density:
Evaluation DimensionObserved Telemetry / AttributeEpistemic Status
Tactical IdentifierAML.TA0001Canonical Tactic Node
Tactic DesignationAdversarial AI & Autonomous Agent CompromiseTactical Dimension
Referential FrameworkMITRE ATLASTaxonomy Anchor
Coverage Index Score89/100Posture Quantification
Defensive Maturity TierADVANCEDOperational Readiness
Monitored Techniques Ratio13 / 15 techniquesCoverage Completeness
Analytical Confidence97%Multi-Sensor Attestation
Primary Evidence SourcesSRC-CISCO-TALOS, SRC-GITHUB-ADVISORIES, SRC-MANDIANT-GOOGLEMulti-Source Grounding

Monitored Techniques & Primary Detection Engines

Section titled “Monitored Techniques & Primary Detection Engines”

The following core techniques within this tactic are systematically monitored by Hermes detection sensors:

Technique IDTechnique NameCoverage LevelPrimary Detection Engine
AML.T0040Direct & Indirect Prompt InjectionHIGHHermes Guardrails / Sigma
AML.T0043AI Agent Tool & Memory PoisoningHIGHChronicle YARA-L
AML.T0048LLM Exfiltration via Markdown & LinksHIGHWAF / Chronicle

Identified operational gaps where adversary tradecraft can evade current sensor placements:

  1. Blind Spot 1: Multi-turn semantic context drift where prompt injection occurs gradually over hours.
  2. Blind Spot 2: Zero-click inter-agent communication spoofing on private WebSocket message brokers.

Prescriptive Detection Engineering Priorities

Section titled “Prescriptive Detection Engineering Priorities”

Hermes recommends the following prioritized implementation roadmaps to close defense gaps:

  1. Priority Action 1: Enforce strict tool parameter JSON schema validation and dual-authorization gates for high-privilege actions.
  2. Priority Action 2: Implement LLM output sanitize filters forbidding unauthenticated markdown remote images.