Blog · ← All posts
Tamper-Seal Certificates: Signed Snapshots That Prove Your Agent Ledger Existed in This Form
By the Ingenious AI team · Published: November 14, 2026 · Reviewed: November 14, 2026 · 7 min read
The short answer: A Tamper-Seal certificate is a signed attestation of an agent ledger's state at a moment in time: the action count, the chain tip hash, and the timestamp, signed with HMAC-SHA256 so anyone holding the certificate can detect if the ledger's history is later rewritten. If the ledger is ever re-presented with a different history for the certified period, its tip and count will not match the certificate — the rewrite is exposed by a single comparison, without re-investigating the records themselves.
Replay a recorded agent incident free → Step through a real recorded run in the replay view, or paste ledger records into the public verifier and get a VERIFIED/BROKEN verdict in seconds — no signup.
A hash chain proves a ledger is internally consistent. A skeptic's next question is sharper: consistent with what? A determined operator could rewrite the entire history and recompute every hash, producing a perfectly self-consistent forgery. The chain alone cannot refute that — it needs an anchor outside itself, a snapshot of 'this is where history stood on this date' that the operator cannot retrospectively alter.
Tamper-Seal certificates are that anchor, built into Agent Flight Recorder as a Pro and Team feature. This guide explains what a certificate contains, how its public verification works, and the operational rhythm — mint at period close, file with your records, present at audits — that makes it evidence rather than decoration.
What is a Tamper-Seal certificate? a cryptographic attestation of a ledger's integrity at a point in time: it certifies one agent's action count and chain tip (the final record's hash), with a minting timestamp, signed using HMAC-SHA256. Anyone can verify a certificate on its public page or JSON API without an account; a ledger whose history was rewritten after certification cannot reproduce the certified tip, and a broken chain can never be certified at all.
What the certificate actually attests
Three facts, no more: this agent's ledger contained exactly N actions at minting time; the chain tip — the hash of record N, which depends on every record before it — was this value; and the chain verified intact at that moment, because certification runs verification first and refuses a broken chain. The economy of the claim is its strength: the certificate does not editorialize about whether the actions were wise, only that this history existed, in this form, at this time.
Because the tip is a function of the whole ordered history, the certificate is effectively a fingerprint of the entire period. Two histories that differ in any record, anywhere, end at different tips with overwhelming probability — so one short string, filed anywhere durable, pins megabytes of history.
Why HMAC, and what verification looks like
The signature uses HMAC-SHA256: a keyed hash that proves the certificate was minted by the recorder service, not fabricated by whoever presents it. Verification is public and deliberately boring: open the certificate's page or its JSON endpoint, and the service recomputes the attestation from the ledger's current state for the certified range and compares. A match says the history presented today is the history certified then. A mismatch says something changed — and the certificate's data (count and tip) bounds the investigation immediately: is the count different (records added or removed) or only the tip (contents altered)?
No account is required for the verifier, and that is a design decision, not an oversight: the party you most need to convince — an auditor, a customer's security team, an opposing expert — should never need your permission to check your evidence.
The operational rhythm: mint at close
Certificates earn their value from rhythm, not from existence. Mint one per agent at every period close that matters to your world: month-end for finance-adjacent agents, quarter close for SOC 2 windows, release boundaries for rapidly changing agents, and — the habit that pays most — at the moment an incident is contained, sealing the history while the events are fresh and before any dispute exists about what the records said.
File certificates with the records they anchor: alongside the verification reports, in the audit packet, in the incident file. A certificate stored only inside the system it certifies is weaker than one also held by a second party — enterprise customers increasingly ask for the certificate at period close precisely so they hold an anchor the vendor cannot rewrite.
What certificates do in an audit or dispute
In an audit, the certificate collapses the integrity interview. Instead of walking the auditor through your storage architecture and access controls for an hour, you present the certificate minted at window close and invite re-verification: if it matches, history is settled and the conversation moves to sampling content. In a dispute, the certificate minted before the dispute existed is the piece that answers 'you could have edited the logs after we complained' — the certified history predates the motive.
There is an honest limit to state plainly: a certificate proves the history from genesis (or the previous anchor) to its minting. It cannot prove events after its timestamp, which is why the rhythm matters — the freshest certificate bounds the unanchored gap, and monthly minting keeps that gap to a month of records that still carry the chain's internal protection.
Five actions you must take today
Anchoring is a calendar habit. Install it like one:
- Mint at every period close, per agent. Month-end at minimum; add quarter closes and incident containment as automatic triggers.
- Verify before you certify. Certification already refuses a broken chain — if a mint fails, that failure is your incident, and it starts now, not at audit time.
- File certificates outside the ledger system. With the audit packet, the board pack, the customer's own records — a second holder doubles the anchor's strength.
- Offer certificates to enterprise customers proactively. A period-close certificate in the customer's hands converts your integrity claim into their own filed evidence.
- Re-verify one old certificate monthly. Pick a past period, re-run the comparison, file the result; anchors that are never checked are assumptions.
Worked example: the certificate that ended an argument
Logistics platform Portside Freight, a fictional Team-plan customer, runs a documentation agent that files customs paperwork. When a shipment was delayed, the broker alleged Portside's agent had submitted a corrected declaration late and then 'the logs were adjusted' to show an earlier time. The allegation was specific, plausible-sounding, and entirely checkable.
Portside's agent had recorded 48,209 actions in the period. The month had closed with a Tamper-Seal certificate — action count 48,209, tip recorded, minted before the shipment dispute existed. Re-verification against the certificate matched exactly, and the declaration's own record sat in the chain at its original sequence with its original timestamp, surrounded by neighbors that also verified.
The argument ended in one exchange of messages: certificate, verification result, and the record. Portside's team noted the operational moral in their incident file — the certificate cost one click at month close and saved a dispute whose legal review alone had been quoted in the thousands.
FAQ
How is a certificate different from just running verification?
Verification checks the ledger against itself, today. A certificate checks the ledger against a signed snapshot from the past. The difference matters exactly when trust in the present state of the system is the question: after an incident, a dispute, or a change of personnel, 'it verifies now' cannot exclude a rewrite that was then re-chained. The certificate can — the rewritten history cannot end at the certified tip.
Can a certificate be forged by the ledger's operator?
The operator holds the ledger but the certificate's HMAC signature is produced by the recorder service's key; an operator of a customer workspace cannot mint a certificate with a backdated timestamp or an invented tip through any supported path, and certificates are publicly inspectable, so an invented one must survive public verification against the live ledger. As with all evidence, the strongest posture pairs certificates with a second holder: file them with customers and auditors as they are minted.
What happens if we purge old records under our retention policy?
Purge whole periods, then mint a fresh certificate for the retained history with the purge documented — date range, policy, authority — in the audit file. The old certificate remains a true statement about its moment; the new anchor re-establishes provability for what remains. Silent mid-history deletion is the one pattern certificates exist to expose, so make every retention action explicit and anchored.
Do certificates prove the actions themselves were legitimate?
No, and they do not claim to. A certificate proves existence and integrity of the history — these records, in this order, unaltered since certification. Whether an action was authorized or wise is judged from the records' contents: approvals, policy versions, reasoning. Integrity evidence and merits evidence are different layers, and confusing them weakens both.
How many certificates should one agent accumulate?
One per meaningful period boundary, kept indefinitely — they are a few hundred bytes each. A three-year enterprise relationship might hold 36 monthly certificates per agent, and that stack is the audit asset: an unbroken series of anchors showing the history was sealed continuously, not reconstructed for the occasion.
Start logging your agents free → The free tier gives you the full hash-chained ledger for 1 agent and 1,000 actions a month, with chain verification and replay included.