SQLGuard · AI Action Authorization · SQL beachhead

AI Production Risk Report

The problem is not only AI mistakes — it is missing proof of what the AI was allowed to do.

band: low · free tip — not authorize

• ok — No write tools / non-prod declared. Re-score when agents gain mutate capability.

Keep PASS+verify on every mutate. Re-run when adding write tools.

Machine: GET/POST /v1/risk · Policy /require · Cite /ai-sql-authorization