CAIN-42 CAIN StudioBlack Box: a flight recorder for AI agents
When an agent refunds the wrong customer, deletes a table or leaks a prompt, the question is "what exactly did it do, and can you prove the log was not edited?" The Black Box records every step as a hash-chained OpenTelemetry gen_ai span, seals the run with an Ed25519-signed RFC 6962 Merkle root, lets you erase personal data later without breaking the proof, and hands an auditor a ZIP they check with no CAIN code.
#10 of the CAIN-42 Top 25 6 gaps found and fixed today 1.1.0 live since 2026-11-02 15:31Z (verified on both gateway instances)
attacks held on today's blind set, before any change
after the fixes
CAIN imports in the auditor's verifier
step can be proven alone, without showing the rest
Try it
2. Tamper with a copy (your browser edits it, the live verifier judges it):
Since 2026-11-02 15:31Z the live server answers NOT VALID to "rewrite the prompt, drop its salt", and erasures report erased_from_disk.
On this site
How it works
Chain, commitments, seal, proofs.
Live demos
The full run plus single calls.
Erasure (GDPR)
What is deleted, what stays, what we found.
Inclusion proofs
Show one step, prove it belongs.
Evidence
23 attacks, every row.
Verify it yourself
The offline auditor kit.
OpenTelemetry
OTLP/JSON export.
API
Routes.
Limits
Honestly.
Use cases
Where it fits.
2026-11-02 upgrade
What we broke and fixed.
FAQ
Questions.
FIRME AI CORPORATION · CAIN-42 is a Byzantine trust fabric. Numbers marked "our own check" were run by us, not by an independent party. Nothing on these pages is a certification.