AI Audit Exchange
Go back

Guide · By the Ingenious AI team · Published 14 November 2026 · Reviewed 14 November 2026

API Keys for Audit Evidence Workflows: A Practical Guide

What is an API key?

An API key is an account credential for programmatic access on AI Audit Exchange. Keys are created through the account's key functions (with owner-issued keys available through the owner path for administration), are verified against stored keyed SHA-256 hashes — the usable secret itself is never stored — and can be revoked, which immediately ends the key's authority. Effective plan features, including key capabilities, follow the account's activated license.

What keys are for in an evidence workflow

Manual monthly filing works at one-room scale; programmes automate. A key lets a script perform the rhythm a human would otherwise click through: scheduled imports staged for review, the shelf's room list and settings pulled into an internal dashboard, manifest exports collected for the annual retention review. The key does not change the evidence rules — imports still seal at intake, badges still record provenance, rooms still bound visibility. Automation inside an evidence system must be more disciplined than the manual work it replaces, because it operates at machine speed: a misconfigured manual import ruins an afternoon; a misconfigured scripted one ruins a period. Build the automation on the same checklist a human follows, and rehearse it on a practice room first.

How keys are stored and why it matters

The platform stores keys the way it stores license keys: as keyed SHA-256 hashes, never as usable secrets, never in URLs, never in logs. Two consequences follow, both healthy. First, a database or log exposure cannot leak working keys — the stored values are hashes, useful for verification and nothing else. Second, a lost key cannot be 'looked up' — not by you, not by support theatre. Loss means revocation and reissue, which is the correct ceremony for any credential. When a vendor offers to retrieve your key for you, they are confessing how they store it. Judge evidence-chain tools by this test early; it predicts much of the rest of their posture.

Issuance discipline: one purpose per key

Issue keys the way you issue grants: per purpose, named, owned. A key for the filing automation, a key for the reporting pull, never one shared key sloshing across teams and laptops. Naming should carry the purpose and owner so the key inventory reads like an org chart of your integrations. Record issuance in the same register as your other evidence-system facts: who holds it, what it touches, when it was created, when it should be reviewed. The register lives outside the room — keys are account-level — but the discipline is the room discipline, applied one level up: enumerate the audience, bound the scope, log the state, revoke without sentimentality.

Rotation and the staff-change rule

Keys outlive the people who created them unless you rule otherwise. Adopt the staff-change rule: when a key's owner leaves the team or changes role, the key is rotated — revoked and reissued to the successor — in the same week as the handover, alongside the room-access review. Rotation is also the answer to any suspicion of exposure, however faint: the ceremony costs minutes and converts a worry into a fact. And rotate on a calendar anyway — annually suits most programmes — so the inventory never contains a key older than everyone's memory of issuing it. A key nobody remembers is functionally identical to a grant nobody remembers: the monthly review should be able to explain every line of both lists.

Revocation and plan reality

Revocation ends a key's authority immediately and is the correct response to loss, suspicion, purpose-end and owner departure. Note the platform's wider key economy, because it teaches the same lesson: license keys activate plans, and a revoked license — after events such as a chargeback on an activated key — returns the account toward free. Credentials throughout the system are states, not possessions: they exist while their conditions hold. Mirror that understanding in your own governance: keys listed as current facts in a register, reviewed like grants, and never treated as permanent infrastructure exempt from the evidence programme's hygiene.

Keeping automation inside the evidence rules

Close with the composition rules that keep scripted work honest. Scripts import through the same paths humans use — suite exports unchanged, uploads badged — because a script that stages 'helpful' transformations has automated the corruption of provenance. Scripts never widen visibility: a key works within its account's rooms, and exports it collects go to governed stores, not inboxes. And script outputs join the review rhythm: the monthly readiness review reads what the automation filed, samples its seals, and checks its rhythm in the manifest's import dates. Automation that survives that review is an asset; automation exempted from it is a liability with good uptime.

Five actions to take

  1. Issue one named key per purpose and owner; keep a register with creation and review dates.
  2. Never share keys across teams, and never embed them where logs or URLs can capture them.
  3. Rotate on every owner change and at least annually; revoke instantly on loss or suspicion.
  4. Automate only through the evidence rules: imports unchanged, badges honest, no widened visibility.
  5. Review key-driven filings in the monthly readiness review — samples, seals and rhythm included.

Worked example

A fictional platform firm, Bright Anvil Systems, automates its filing with two keys: 'filing-automation — owned by compliance ops' staging monthly imports for human confirmation before they land, and 'reporting-pull — owned by the audit programme lead' collecting room lists and manifest exports for the annual review. The register fits on half a page. When the programme lead changes roles in October, both her key's rotation and the grants review happen in the same sitting — twelve minutes — because the SOP names them as one ceremony.

In the following year's review, the key inventory reads: two active keys, both owned, both rotated within twelve months, one revoked experiment key from the programme's first month shown as historical. The auditor of the current cycle, reading the register as part of the control environment, notes the sentence the whole exercise exists to produce: 'Credentials in this organisation are enumerated, bounded and reviewed — including the machine ones.'

Frequently asked questions

Where is my API key stored?

Only as a keyed SHA-256 hash. The platform can verify a presented key against the hash but cannot retrieve the usable secret — loss means revocation and reissue, by design.

Can a key access other accounts' rooms?

No. Keys act within their own account's authority; room visibility remains owner-plus-granted-auditor. A key extends your tooling, never your audience.

How many keys should we hold?

As few as purposes require — typically a small number, each named and owned. Every additional key is an additional line the monthly review must be able to explain.

What happens when a key's owner leaves?

Rotate in the handover week: revoke the old key, reissue to the successor, update the register. Pair it with the room grants review; they are the same ceremony at different layers.

Do keys bypass the evidence rules?

No. Key-driven imports seal, badge and manifest identically to manual ones. Automation changes who clicks, never what counts as evidence handling.

Keep reading

All AI Audit Exchange guides · Auditor directory · About AI Audit Exchange · Security · Pricing · AI Audit Exchange home · Previous: Evidence Room Security: How Access and Accounts Are Controlled · Next: How to Hand Over Evidence to a New Auditor

Version Control for Audit Evidence Without Editing the Past · What Is Source Badging on Audit Evidence? · Importing ProvenanceGuard Evidence Into an Audit Room