
Two questions every auditor now asks
When the auditors arrive, they ask two things about your AI agents: what did they do, and can you prove nobody changed it?
- Today the trail is scattered across application logs, screenshots and spreadsheets.
- Pulling it together takes engineers off their real work.
- And at the end, the auditor still has to take your word that nothing was edited.
Evidence an auditor cannot test is a statement, not evidence.
A signed record of every decision
Argent Strategic Intelligence sits between your AI agents and the AI models and tools they use. Every request that passes through it leaves a record.
- Signed and linked. Each record carries its own digital signature and is linked to the record before it, so a changed or removed record shows.
- Recorded or refused. A record that cannot be signed is not written, and by default Argent refuses new requests when it cannot take their records, rather than serve them unrecorded.
- Sealed every day. At midnight UTC each day is sealed with a signed daily root that links it to the next day.
- Admin actions too. Changes made in the Argent console go on the same signed record.
One click builds the package
The auditor does it themselves, with read-only access. Nobody in engineering has to pull a thing.
- Open Evidence. In the Argent console, the auditor opens Evidence, then the Exports tab.
- Pick a period. Any past period of whole, sealed days, up to 31 days per package, for the whole company or for one agent.
- See what it will hold. The preview fills in on its own: traces, rows and the list of files, plus an estimated size once an earlier package exists.
- Click BUILD PACKAGE. Before a single byte leaves, taking the copy is recorded in the signed log under the auditor's name. If that record cannot be written, nothing is exported.
- One file downloads. A ZIP of the company's own signed records, the public keys, a signed manifest that lists every record it holds, a README that explains how to check it, and a SHA-256 checksum for every file.
The auditor checks it on their own machine
Every auditor's first objection: couldn't the vendor just hand over a clean file? So the check does not run on Argent.
- Confirm the key. The auditor matches the package's signing key against one their firm already holds, from a source other than the package. The verifier never takes its trust key from the package itself.
- Run the verifier. On their own laptop, with no connection to Argent and no Argent account, they run
offline-verifier verify-range. It checks every record's signature, every day file and the signed manifest. - Read the answer. A sound company package prints
TENANT PACKAGE OK. A missing, extra, reordered, duplicated or altered record printsTENANT PACKAGE FAILEDand names the file. A one-agent package proves every record in it belongs to that agent.
Checked by the auditor, not taken on your word.
What the operator sees in the console
Most of it lives on one page, Evidence, with the same signed record behind every tab. The rest sits on the agent page, Reports and Alerts.
| Where | What it shows |
|---|---|
| Evidence: Decisions | Every decision in the period, with EXPORT CSV. The file's SHA-256 is recorded in the signed log before the first byte. |
| Evidence: Events | The signed rows exactly as written, exportable unchanged as NDJSON. |
| Evidence: Integrity | A verdict per period, INTACT, BROKEN or VERIFIED, with the checks behind it: continuity, signatures, anchor and last pass. VERIFIED means the signing keys are also pinned. |
| Evidence: Exports | BUILD A PACKAGE, and COPIES TAKEN: who took each copy, when, the scope, rows, size and SHA-256. |
| Agent page: Audit | BUILD EVIDENCE PACK, a package of one agent's records. |
| Reports: Evidence | A platform administrator's package for the whole install, with a suggested mapping from SOC 2 and ISO 27001 criteria to the records that answer them. |
| Alerts | Chain breaks, missed verification passes and key mismatches, sent to the console, a webhook or email. |
What auditors and regulators already expect
The rules you already answer to ask for a record of system activity you can stand behind.
- Healthcare, HIPAA Security Rule, 45 CFR 164.312: covered entities must "implement hardware, software, and/or procedural mechanisms that record and examine activity in information systems that contain or use electronic protected health information," and protect that information "from improper alteration or destruction."¹
- New York financial services, NYDFS 23 NYCRR 500.6: covered entities must, to the extent applicable and based on their risk assessment, keep systems that "include audit trails designed to detect and respond to cybersecurity events" that could materially harm their operations, and keep those audit-trail records "for not fewer than three years."²
- SOC 2 and ISO 27001 audits: Argent's Reports page maps your records to these criteria as a suggested reading guide for your auditor.
Where it fits in your program
Built for the teams that answer to auditors: security, risk, compliance and internal audit.
- Goals: Audit and accountability, Compliance evidence, See all AI in your company and Sovereign AI.
- Industries: Financial services, Insurance, Healthcare, Public sector and Legal, or see all industries.
- Use case: AI in regulated and air-gapped environments.
- Pairs with: Session Replay to follow one session prompt by prompt, Watch Before You Enforce to start the record before anything is blocked, Identity Tethering so every agent has a named human owner, and Give the Board What It Wants. See all capabilities.
- Deployment: air-gapped on premises, private cloud or your own public cloud VPC; compare options.