Tamper-evident audit log demo
Verify a sample audit chain yourself
Our core claim is that agent audit evidence should be independently verifiable — by you, not by our word. Most audit logs answer one question: valid, yes or no? Ours names the crime: tampering, insertion, deletion, or truncation — each a distinct, reportable verdict. No signup, no sales call.
Want to break it yourself?
The interactive sabotage demo — tamper, insert, delete, or truncate a live hash chain in your browser and watch verification name the crime in real time — lives in the Demo Center, alongside break-glass, billing-verification, key-rotation, and agent-drift demos. Entry is a quick work email and a mutual NDA — you get a secure magic link in seconds.
What verification sees
Verification is classification, not a yes/no: each kind of damage produces a distinct, reportable verdict.
How hash chain verification works
Every audit entry carries a SHA-256 hash of its own content plus the previous entry's hash — alter any row and every hash after it breaks. Three properties make the log tamper-evident, not merely tamper-resistant:
Hash-linked rows
Each entry commits to the one before it. A single edited field invalidates every downstream hash — you cannot quietly rewrite history.
Signed checkpoints
At intervals an Ed25519-signed checkpoint seals the chain state, and checkpoints are themselves chained — so a deleted checkpoint is detectable, never silent.
Fails closed
Rotation is kid-keyed: a checkpoint signed by an unknown key is a verification failure, not a warning. No key, no trust.
Why this matters for EU AI Act Article 12 and HIPAA
A log an administrator can silently edit has no evidentiary weight. This chain is designed to map to the obligations that put you on the hook — what you retain is a structure a regulator, or you, can re-verify at any time. And one property matters most in regulated deployments: GDPR erasure doesn't break it — actor identities are HMAC-committed, so crypto-shredding a data subject removes re-identification while every hash still links.
EU AI Act — Art. 12
Lifetime event logging for high-risk AI systems.
EU AI Act — Art. 26
Deployers retain logs at least six months — the obligation is yours, not your model vendor's.
HIPAA — §164.312(b)
Audit controls over ePHI access — and AI agents touching ePHI are inside that requirement.
Erasure destroys the key, not the evidence
The question every SIEM fails: how does GDPR Article 17 erasure interact with an immutable log? Crypto-shred resolves the tension — destroy the per-subject key and raw values become unrecoverable, while HMAC-committed actor identities keep every hash linking.
Erasure and evidence integrity, simultaneously — the property one-way redaction vendors and mutable-log platforms structurally cannot offer.
The verification report, verbatim
The audit_verify_chain operation returns a structured report — the same schema whether we run it or you do. For a walkthrough against your own deployment, ask the quoting agent.
| Field | What it returns |
|---|---|
status | VALID, or the failure classification |
entriesChecked · checkpointsChecked | coverage counts |
firstBrokenSeq | where integrity fails, if it does |
gapClassification | tampering vs. insertion vs. deletion vs. truncation |
unsignedTail | entries after the last checkpoint — a truncation window, reported honestly |
attestedDamage | previously acknowledged damage (ISO 27001 register) that can never mask fresh tampering |
Want this chain under your agents?
Get an instant quoteYou've seen the proof
Ready for a number? Scope your deployment and we'll price it against your own economics.
Get your quote →