Every AI call your team makes — every prompt, every multi-agent run, every Cursor session — passes a policy gate, lands in a tamper-evident audit log, and exports as a signed Evidence Pack your auditor can verify offline.
Built for CISOs and Risk Officers who need to defend AI use cases before a regulator does.
Every AI decision lands in a hash-chained log keyed by org_id
so cross-tenant tampering is mathematically detectable. Export as a signed
Evidence Pack — the tools/verify_evidence_pack.py CLI ships
with the bundle so your auditor verifies offline, no backend access.
NIST AI RMF · ISO 42001 · EU AI Act — 59 baseline controls pre-mapped
per skill, with per-org overrides. Closed-set hallucination guard cross-checks
every cited control ID against a 259-entry truth set: a fabricated
NIST AI RMF GV-9.99 can't pass.
Per-org monthly USD cap with audited refusal — over budget, the LLM call never fires. Four hallucination signals run on every output: closed-set membership, LLM-as-judge grounding, semantic-entropy consensus, cross-vendor disagreement.
Proofpane learns from its own audit log — the dreamer loop
proposes prompt upgrades against drift / hallucination signals, the
sandbox replays them on past failing cases, and only the changes a
human approves go live. Every proposed and approved change is signed
into the same hash chain — never a black-box silent improvement.