Framework lens
EU AI Act — documentation support
A documentation and evidence coverage lens for the EU AI Act — coverage of expectations, never a legal verdict.
A documentation and evidence coverage lens for the EU AI Act — coverage of expectations, never a legal verdict.
Why it matters
Teams shipping AI agents that touch people in the EU want to know what the EU AI Act expects them to be able to show, and where their evidence is thin, before a stakeholder asks. A documentation lens organises that and prepares structured information for your own legal and risk review — it does not stand in for it.
What good looks like
The same agent viewed through the EU AI Act lens as a coverage map, not a grade. A high-impact use — for example biometric or emotion recognition, employment decisions, or essential-services eligibility — raises a high-risk candidate flag and a recommendation for legal review, framed as 'may apply' rather than a determination. Unknowns stay visible and can hold applicability at 'Not enough information'; an unknown on a high-impact trigger conservatively escalates rather than being read as no. Evidence expectations map back to the evidence register and controls.
What can go wrong
A tool claims an agent has passed the Act or is fully aligned with it — a legal determination no documentation product can make. Unknowns are silently treated as not applicable. A high-impact case is missed because triage never flagged it. The mapping is mistaken for legal advice.
What AgentProof checks
AgentProof runs one deterministic, seed-driven lens engine across all profiles; the EU AI Act differences live in seed data pinned to a named framework version. It produces documentation and evidence coverage plus an applicability read (may apply, likely relevant, not enough information, not currently relevant) and fires legal-review flags on high-impact domains. It is coverage-not-compliance and reads-not-rescore: it never recomputes the readiness score and makes no legal conclusion. Your legal, privacy, and risk teams remain responsible. AgentProof does not provide certification, an audit verdict, or legal advice.
Key terms
The exact vocabulary this part of the record uses — grounded in the shipped product model.
- Applicability
- may_apply, likely_relevant, not_enough_information, not_currently_relevant.
- Review flags
- legal, compliance, security, privacy — raised where a high-impact trigger is present.
- Quick-triage tiers
- light, standard, enhanced, strict — a high-impact case escalates to strict.
- Framework version
- The EU AI Act seed is pinned to a named framework version.
- Unknown never hidden
- An unknown on a high-impact trigger escalates rather than being read as not applicable.
Keep reading
Build this record for your own agents
AgentProof turns each part of this into one documented, evidence-backed record — before your compliance review.
Ready to build a compliance-readiness record for your own agent?
Request a compliance readiness pilot to apply this guidance to a real agent.
AgentProof builds a compliance-readiness record, not an official audit, and it does not speak on behalf of any vendor.