Shield Agent AI

← Docs hub

Customer guide

EU AI Act crosswalk

kya certify evaluates the open Agent Trust Baseline (30 requirements across Data and Privacy, Security, Safety, Reliability, Accountability, Society) against local evidence and writes a gap report (JSON + Markdown + HTML) under .kya/certify/.

kya certify            # gap report; exit 1 when gaps exist (CI-friendly)
kya certify --sign     # also emit an ed25519-signed evidence-bundle.json
  • Every requirement maps to EU AI Act articles 9 to 15, 26, and 72 in docs/certify-eu-ai-act.md and src/certify/eu-ai-act-map.json.
  • Machine checks never pass on an empty trail; they report insufficient_evidence.
  • What a machine cannot check is covered by explicit local attestations, labeled as unverified operator statements.
  • Certify is evidence-only: it never ALLOWs, DENYs, or blocks. The sole policy enforcement point stays the gate.

Bundles from kya certify --sign are self-signed. They prove bundle integrity and key continuity, not identity. They show a grey self-signed badge on the public verify endpoint (POST /api/v1/kya/evidence/verify) and badge. The green Shield Verified badge is only for bundles signed with the key of an active hosted workspace. A green badge needs a bundle that was downloaded on Pro or Scale and is not older than 7 days. It ties the badge to a hosted account, not to a checked legal identity of the company or the agent. It is not a third-party audit.

Hosted certify

The hosted console evaluates the same 30 requirements against your hosted trail: /app/kya/certify shows the live requirement table with evidence and recommended actions, and attestations can be recorded in the page. GET /api/v1/kya/certify/report returns the gap report on every plan; POST /api/v1/kya/certify/bundle downloads an ed25519-signed evidence bundle on Pro and Scale. Hosted bundles use the OSS bundle format byte-for-byte, so the public verify endpoint validates both without changes.

Docs hub · Install · How KYA works · Source