Guide · By the Ingenious AI team · Published 14 November 2026 · Reviewed 14 November 2026
Evidence Room Security: How Access and Accounts Are Controlled
Short answer
An evidence room's security rests on layered controls the platform states plainly: room contents are visible only to the room owner and to an auditor the owner has granted that room; artifacts are SHA-256 sealed at intake and stored immutably, with manifest hashes re-verifiable at any time; sessions use signed, HttpOnly, SameSite cookies; API keys are stored only as keyed SHA-256 hashes and never appear in URLs or logs; authentication endpoints are rate-limited; and the room keeps an access log. No single layer is the story — the composition is.
What is room security?
Room security is the set of platform controls protecting evidence rooms on AI Audit Exchange, described on the platform's security page: hash sealing and immutability for artifacts, visibility restricted to owner and granted auditor, signed session cookies, keyed-hash storage for license and API keys, parameterised queries, rate-limited authentication endpoints, credential-gated owner administration, and per-room access logging.
Visibility: the smallest possible audience
The foundational control is audience size. Room contents are visible only to the room owner and to an auditor the owner has granted scoped access to that room — two categories, no ambient audience of colleagues, administrators-by-default or other customers. Every other control protects a boundary; this one decides where the boundary is, and it is drawn tight enough that the access log stays human-readable. When you evaluate any evidence system, ask its version of this question first: who can see this, enumerated? If the enumeration is long or vague, the remaining controls are decoration on an open door. Here the enumeration is the owner, plus named grants, each revocable, each logged.
Sessions and accounts: the identity layer
Access rides on accounts whose sessions use signed, HttpOnly, SameSite cookies — the standard trio that keeps session tokens out of scripts and cross-site requests. Authentication endpoints are rate-limited, blunting credential guessing. Secrets are handled as evidence systems should handle them: license keys and API keys are stored only as hashes — API keys as keyed SHA-256 — and never appear in URLs or logs, so a leaked log line or a shared link cannot leak a key. Owner administration is credential-gated separately. None of this is exotic; that is the point. Security postures fail on exotic cleverness and succeed on standard controls, completely applied.
Artifact integrity: sealing and immutability
Inside the room, the artifact layer carries its own controls independent of access: every artifact SHA-256 sealed at intake, stored immutably, its hash printed in the manifest for re-verification by owner or granted auditor at any time. Even a hypothetical reader who reached a file they should not have cannot quietly alter the record — the seal would testify against the altered copy anywhere it surfaced. Imports arrive read-only and unchanged from their producing products, which closes the supply chain too: the file the product produced is the file the room holds is the file the auditor reads, with a computation proving the chain at every link. The mechanics are in SHA-256 sealing explained.
The access log: security that can be shown
Controls you cannot exhibit are promises; the room's access log converts access control into an exhibitable fact. The log records the room's accesses, and the room page lists its active grants with the auditor's firm name, so the owner can answer 'who has seen this evidence?' from the room itself — at acceptance, mid-engagement, at close-out, or in year four when someone asks historically. Build the log into your review rhythm: the readiness review's access segment exists because the log does. An auditor reading a room whose log shows disciplined, expected access — and whose grants list is exactly as long as the engagement requires — prices the whole control environment upward. See sharing evidence securely for the model in full.
Guests, keys and the edges of the model
Two edges deserve plain statement. Guest mode: the free room runs with no sign-in through a guest account context, which is how the front door stays open; guests operate their own room's evidence pattern, and account-grade capabilities — engagements, API keys, plan shelves — belong to durable identities. API keys: where accounts create keys for programmatic access, keys are stored as keyed hashes and shown under the account's controls; a key is a credential for its owner only, and the platform's recovery answer to a lost key is revocation and reissue, never retrieval of the stored value, because the stored value is a hash. Edges stated plainly are edges that hold.
Reading a security page like an auditor
This platform publishes its posture on its security page, and the readiness habit generalises: read every tool in your evidence chain the way an auditor reads you. Claims sorted into checkable and uncheckable; the checkable ones — hash verification above all — actually checked by your own hands in rehearsal. A vendor whose central integrity claim is personally testable by the customer is offering a different species of assurance from one whose claims terminate in a badge image. Choose accordingly, and hold your own organisation's claims to the same standard in your rooms: manifest, seals, logs — the facts you would want if you were examining you.
Five actions to take
- Enumerate your room's audience monthly: owner plus named grants only — revoke anything else on sight.
- Keep authentication hygiene boring: strong account credentials, keys treated as credentials, lost keys revoked and reissued.
- Verify seals personally in rehearsal; never rely on an integrity claim you have not recomputed.
- Read the access log at acceptance, mid-engagement and close-out, and keep it in the readiness review.
- Judge every tool in your evidence chain by the same standard: checkable claims, checked.
Worked example
A fictional B2B platform, Cobalt Works Systems, puts its own CPPA room through a security self-examination before engaging its auditor. The enumeration takes four minutes: owner account, one grant (the auditor of record, created at acceptance), access log showing the auditor's 23 reads across 5 sessions and no one else's. Five sampled artifacts verify against their manifest seals. The account's API key inventory is reviewed: one active key, stored (as designed) only as a hash, owned by the integration that uses it.
One finding emerges, small and exactly the kind the exercise exists for: a second API key, created during an early experiment and forgotten, is still active. It is revoked in the sitting. The room's note records the whole examination in six lines. When the auditor later asks how access to the evidence is controlled, the answer is not a policy PDF — it is the log, the grants list, the seals and that six-line note, all inside the room she is already reading.
Frequently asked questions
Who can see my room's contents?
Only you, the room owner, and an auditor you have accepted and granted that room. There is no broader visibility — not other customers, not other rooms' auditors.
How are API keys protected?
Keys are stored only as keyed SHA-256 hashes and never appear in URLs or logs. A lost key is revoked and reissued; the stored value cannot be retrieved because it is never stored in usable form.
What protects against brute-force sign-in attempts?
Authentication endpoints are rate-limited, and sessions use signed, HttpOnly, SameSite cookies. Owner administration is separately credential-gated.
Is the access log visible to the auditor too?
The room's working record — including its access information — is part of how the room demonstrates its controls within the engagement's scope. Both sides benefit from access being a shown fact rather than an asserted one.
Does guest mode weaken room security?
Guest mode operates the free room through a guest account context so the evidence pattern runs with no sign-in. Artifact sealing, immutability and room-level visibility apply in guest rooms as elsewhere; account-grade capabilities require a durable account.
Keep reading
All AI Audit Exchange guides · Auditor directory · About AI Audit Exchange · Security · Pricing · AI Audit Exchange home · Previous: How to Run an Internal AI Audit Readiness Review · Next: API Keys for Audit Evidence Workflows: A Practical Guide
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?