MITRE ATLAS

Establish and maintain documented provenance and lineage for all training, fine-tuning, and evaluation data used by AI systems, including legal basis, licensing, sensitivity classification, and transformations applied.

Validate that production-bound LLM and agentic AI systems have been tested against direct prompt injection, indirect prompt injection, jailbreak, refusal evasion, and (where applicable) multi-modal adversarial inputs, with documented findings, remediation, and re-test cycles.

Ensure that third-party AI products and AI-enabled services are subject to risk-based due diligence covering data handling, model provenance, security testing, incident response, compliance posture, and contractual safeguards before procurement and on an ongoing basis.

Apply layered output filtering and content moderation to LLM and generative AI systems to prevent disclosure of sensitive data, prohibited content, executable payloads, and policy-violating outputs.

Ensure that AI systems produce logs sufficient to detect abuse, investigate incidents, demonstrate compliance, and reconstruct decisions, with logs protected from tampering and retained per regulatory requirements.

Ensure that every model promoted to production is uniquely versioned, traceable to its training data, evaluation results, and approval, and subject to controlled change-management with rollback capability.

Ensure that personal data used in training, fine-tuning, evaluation, and inference is minimized to the data strictly necessary for the stated purpose, processed under a defensible lawful basis, and subject to the rights of affected individuals.

Validate that production predictive ML, computer-vision, recommender, and speech models have been tested against evasion attacks, model-extraction attacks, membership-inference attacks, data-poisoning detection, and (for vision) physical-world perturbations, with documented findings and remediation.

Restrict agentic AI systems to least-privilege tool authorizations; enforce approval boundaries for high-impact actions; log and review all tool invocations; and prevent privilege escalation across multi-step or multi-agent workflows.

Detect, triage, and remediate use of unsanctioned AI services and unauthorized AI tooling by employees, contractors, and other authorized users, with documented response and metrics.