
ARGENT FORCOMPLIANCE EVIDENCE
Master Control System · by Nyedis
Your goal: prove your AI is under control
If your goal is to show auditors and regulators that your AI is under control, a policy on paper is not enough. They ask for the record that shows the policy ran.
- IBM found 63% of breached organizations either had no AI governance policy or were still developing one.¹
- Of the organizations that had a policy, only 34% performed regular audits for unsanctioned AI.¹
- Of the organizations that reported a breach of an AI model or application, 97% had no AI access controls in place.¹
A policy is a promise. Evidence is the record that the promise was kept. Argent writes that record on every request that goes through it, and hands it to your auditor in a form they can check.
Five questions an auditor asks about your AI
Each question has an answer in the Argent console, backed by the same signed record.
| The auditor asks | Where the evidence comes from |
|---|---|
| Which AI agents do you run, and who answers for each one? | The Agent inventory. Every registered agent has an active, named human owner, or its requests are refused. Each agent's history lists every edit, key event, owner change and quarantine, and a retired agent is archived, never deleted. |
| Who may do what, and does anyone review it? | The access review on the Entitlements tab lists every scope grant, role and group membership with its last Keep decision and when the next one is due. Each Keep is an append-only certification, dormant agents are flagged for revoke, and the review downloads as CSV. |
| Which policies were in force, and when? | The Policy register shows nine kinds of policy in one list, each with its history, exportable as CSV. Signed events show which controls were in watch and which in enforce at any moment. A change to a sensitive-data policy is versioned and signed before it lands. |
| What did the controls actually do? | The Reports page counts refusals by policy, such as sensitive data, prompt injection, unpriced models and missing owners. Sensitive-data records hold the kind and count, never the value. The retrieval log shows who retrieved what, and what was withheld. |
| Can you prove nobody changed the record? | Every record is signed and linked to the one before it, and each day is sealed with a signed daily root. Your auditor checks it on their own machine. See One-Click Audit Evidence. |
The path: from first request to evidence package
Evidence builds up as you go. The record starts on the first request, before anything is blocked.
- Route your AI traffic through Argent, in watch mode. Requests go through as normal, and what each control would have refused or changed goes on the signed record. See Watch Before You Enforce.
- Give every agent a named owner. A developer registers each agent with an active human owner, and when that person leaves, the agent is refused on its next request. See Identity Tethering and Owner Offboarding.
- Set your policies and test them first. Test a policy against one agent's real access, which writes only a simulated record, then switch controls to enforce one at a time. Each audited admin change is signed in the same transaction as the change itself.
- Review access on a schedule. Keep or revoke each grant on the Entitlements tab, and attach the CSV to your compliance ticket.
- Read your posture. Reports shows eight checks against the previous period, each with a plain-words explanation and a link to the records behind it, and prints as a board summary. See Give the Board What It Wants.
- Hand the auditor a package. An Auditor account opens Evidence, picks a past period of up to 31 sealed days and clicks BUILD PACKAGE. The auditor checks it on their own laptop.
Evidence your auditor checks themselves
Every auditor's first objection: couldn't the vendor just hand over a clean file? So the check does not run on Argent.
- A read-only Auditor role. The auditor reads Evidence, Reports, Live traffic, Sessions and replay, the access review and the agent list, and cannot change anything.
- One click builds the package. A ZIP of your 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.
- Checked on their own machine. The auditor matches the signing key against one their firm already holds, then runs
offline-verifier verify-rangewith no connection to Argent. A sound package printsTENANT PACKAGE OK; a missing, extra, reordered or altered record printsTENANT PACKAGE FAILEDand names the file. - Every copy is on the record. Before a byte leaves, the export is signed under the auditor's name. COPIES TAKEN lists who took each copy, when, the scope, rows, size and SHA-256.
- One request, end to end. A platform administrator can export one request's signed rows with the keys to check them. Follow any record to its full Session Replay.
A record that tells the truth about itself
Evidence is only worth something if its gaps show. Argent is built to say what is missing rather than look clean.
| What could go wrong | What the record does |
|---|---|
| A record cannot be signed | It is not written, and by default Argent refuses new requests rather than serve one it cannot record. |
| An accepted event is lost before it is written | The chain carries a signed gap marker with the count and the time span, and the offline verifier reports it. |
| Someone explains a gap | A platform administrator can attach a signed reason beside the gap. It never changes the day's status or the verdict. |
| The chain breaks or the checker stops | The Integrity tab shows a verdict per day, and anyone who subscribes gets an alert in the console, by webhook or by email when a break is found or a scheduled check is missed. |
Kept for as long as your rules require. When your operator connects the write-once archive, each sealed day is uploaded under object lock, seven years by default, and Argent refuses to start with a weaker setting unless the operator explicitly allows it. Database copies are dropped only after that day is archived, fetched back and verified. Captured prompts and replies sit in a separate encrypted store, kept until you set a retention period.
What frameworks and regulators already ask for
The rules you already answer to ask for documented controls and a record of system activity you can stand behind.
- Any organization using the NIST AI Risk Management Framework: GOVERN 1.4 asks that "the risk management process and its outcomes are established through transparent policies, procedures, and other controls based on organizational risk priorities," and GOVERN 1.5 that "ongoing monitoring and periodic review of the risk management process and its outcomes are planned, organizational roles and responsibilities are clearly defined."²
- 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."³
- New York financial services, NYDFS 23 NYCRR 500.6: covered entities must keep systems that "include audit trails designed to detect and respond to cybersecurity events," and keep those audit-trail records "for not fewer than three years."⁴
- SOC 2 and ISO 27001 audits: the Evidence tab on Argent's Reports page maps each criterion to the Argent record that answers it, as a suggested reading guide for your auditor.
Argent supplies the records. Your auditor reads them against the framework you answer to.
Where to go next
Compliance evidence draws on every other control. These pages take it further.
- Capabilities on this path: One-Click Audit Evidence, Watch Before You Enforce, Identity Tethering, Owner Offboarding, Session Replay, Give the Board What It Wants and Sensitive Data Redaction. All capabilities.
- Related goals: audit and accountability, see all AI in your company, data loss prevention, agent identity and ownership, model control and routing, budget and token spend, threat defense and sovereign AI. All goals.
- Use cases: AI in regulated and air-gapped environments, Securing AI coding agents, Governing customer-facing AI assistants, Rolling out copilots and chat safely and Controlling agent tool access.
- Industries: Financial Services, Insurance, Healthcare, Public Sector, Legal and all industries.
- Where it runs: inside your own environment: air-gapped on-premises, private cloud or your public cloud VPC. Compare deployment options.
- See it yourself: take the walkthrough, download the data sheets or browse our integrations.