EU AI Act
Article 14
Human oversight evidence for high-risk AI systems
What the regulation requires
Article 14 requires deployers of high-risk AI systems to implement human oversight measures. Providers must design systems to enable oversight through mechanisms built into the user interface or made available to deployers. Oversight must prevent or minimize risks to health, safety, and fundamental rights.
How RADAR maps to it
Decision logging
Every AI agent decision is captured with full context — input, output, model, timestamp. Each record is hashed with SHA-256 and linked into a Merkle chain that an external auditor can verify independently.
Oversight intervention
Policy engine with kill-switch capability. When a decision falls outside defined parameters, RADAR flags it and can halt the agent pipeline. Every intervention is logged as an evidence event.
Transparency records
Evidence packs exportable in PDF, JSON, HTML, CSV, JSONL, and Parquet. Each format includes full audit trail with chain verification status and auditor-verifiable hash references.
Accuracy monitoring
Continuous monitoring of agent outputs with drift detection. Changes in decision patterns are flagged and timestamped. Evidence covers the entire operational period — not just snapshots.
What the evidence looks like
Auditor note
Verifiable without vendor access. Auditor validates hash chain independently.
This is a production-verified capability — not a hypothetical. Every control mapping above corresponds to a feature shipped in RADAR v1.0. See the documentation.