Guide · By the Ingenious AI team · Published 14 November 2026 · Reviewed 14 November 2026
How to Run an Internal AI Audit Readiness Review
Short answer
Run an internal readiness review from the evidence room outward, in one sitting: read the manifest's coverage, sample-verify hashes, work the validation stale list, count the completeness steps, inspect grants and the access log, and drill five auditor questions in the room's thread. Every output is a dated fact or a named task — never a vibe. Quarterly in the run-up to an engagement, monthly in its final stretch, the review converts audit anxiety into a shrinking list of specific, ownable fixes.
What is a readiness review?
An internal readiness review is the auditee's structured self-examination of an evidence room before and during an audit cycle. On AI Audit Exchange it is assembled from the room's own instruments: the exported manifest, per-artifact hash verification, validation with the 365-day freshness rule, the five-step completeness view, the grants list and access log, and the question-and-answer thread used as a drill space.
The review's one rule: facts or tasks, nothing else
A readiness review fails when it produces reassurance. Its outputs must be exclusively facts ('41 artifacts; sources complete for Q1–Q3; 3 of 3 sampled hashes verified') and tasks ('import October's CompanionGuard exports — owner named, due Friday'). Anything that is neither — impressions, confidence levels, general satisfaction — is decoration that will dissolve on contact with an actual auditor. Structure the sitting so decoration has nowhere to sit: the agenda below is a sequence of readings from the room's own instruments, each ending in a number or a name. If your last review produced no tasks, it was not a review; it was a ceremony. Real rooms always have a next filing, a next flag, a next question.
Agenda part one: the manifest passes
Open with the exported manifest — re-export it fresh so the review reads the current room. Coverage pass: sources present per month of the period; import rhythm steady or clustered; direct-upload share sane. Integrity pass: three to five artifacts sampled the auditor's way — consequential, old, uploaded, twinned — each hash verified against its seal and the result stated in the review note. Administration pass: regime, period, retention months and hold status read aloud from the room block. Fifteen minutes, three passes, and the room has told you more truth than most status meetings produce in a quarter. The passes are detailed in exporting the manifest.
Agenda part two: freshness and completeness
Run validation, or read the latest run, and work the stale list as a filing list: which flagged artifacts gain current-period equivalents, who imports them, by when; which are legitimately historical and get their acknowledgement noted in the thread. Then read the completeness view and name the missing steps precisely — engagement not yet posted, manifest export older than the last filing round, retention still at default. Each missing step converts to a task with an owner in the room. The trend matters more than the snapshot: stale count falling across reviews, completeness climbing toward 5 of 5 as the period matures. A review that tracks trends manages a programme; one that reads snapshots manages surprises.
Agenda part three: access and questions drill
Inspect the grants list and skim the access log: exactly the auditors who should hold access, holding it; reads consistent with the engagement's rhythm; no orphaned grants from dead conversations. Revoke on the spot anything without a live justification — the review is the natural home of the revocation habit. Then the drill: five questions, written in the room's thread as an auditor would write them — anchored to manifest entries, absences and validation results — and answered from the room by the people who will answer the real ones. Questions that cannot be answered become the review's most valuable tasks, because they were found by friends, in rehearsal, with the calendar still kind. The workflow is described in questions and answers in the room.
Who sits in the review, and how long it takes
Keep it small and senior enough to own tasks: the room's owner, the person who files, and one sceptic whose job is sampling and awkward questions — rotate the sceptic so the role stays honest. Ninety minutes suffices for a single room; multi-room programmes review room by room and resist blending the numbers. Hold the review in the room's own surfaces — screens shared from the manifest, the validation view, the thread — rather than in a slide deck about them. Slides describe; rooms testify. The note that leaves the meeting is one page: facts read, tasks named, trends stated. File it as a direct upload into the room it reviews, where it becomes part of the record it describes — a habit auditors read with visible pleasure.
Cadence: quarterly, then monthly, then once more
Run the review quarterly while the period is young, monthly across its final stretch, and once more in the week before the auditor's start date, as the last rehearsal. Between reviews, the monthly filing rhythm continues untouched — the review judges the rhythm, it does not replace it. Over a full cycle, the review notes accumulate into the programme's own evidence: a dated series showing coverage maintained, integrity sampled, flags worked, access governed. When the auditor asks the meta-question — 'how do you know you are ready?' — the honest answer is that series. Readiness was never a state you declared; it was a practice you can show.
Five actions to take
- Run the review from the room's instruments only: manifest, verification, validation, completeness, grants, thread.
- Produce facts and tasks exclusively — every line of the note is a number read or an owner named.
- Sample hashes the auditor's way: consequential, old, uploaded and twinned artifacts included.
- Drill five anchored questions in the thread and convert unanswerable ones into tasks immediately.
- Meet quarterly, then monthly near the engagement, and file each one-page review note into the room it reviewed.
Worked example
A fictional consumer-services platform, Velvet Antler Services, runs its third quarterly readiness review on a CPPA room. The note's facts: 58 artifacts, four sources complete through the quarter; 4 sampled hashes, 4 verified; stale count down from 9 to 3 across reviews; completeness 4 of 5 — retention set, manifest current, engagement posted, auditor matched, with 'evidence filed' long since true and re-earned monthly. Access review finds one orphaned grant from a preliminary auditor conversation in spring; revoked in the sitting.
The drill produces the meeting's real yield: question three — 'which artifact evidences the June configuration change?' — cannot be answered from the room. The change record exists in the producing product but was never imported. It is imported within the hour, and the answer is posted in the thread pointing at the new manifest entry. One question, one gap, one fix, zero findings when the real auditor asks the same question eleven weeks later — because she does, almost word for word. Rehearsal is only prophecy when the questions are honest.
Frequently asked questions
How is a readiness review different from the audit itself?
Ownership and consequence: the review is yours, run by friends, and its findings are tasks. The audit is the independent examination. The review's purpose is to ensure the audit finds a room whose owners already interrogated it harder.
Who should play sceptic?
Rotate the role among people close enough to the evidence to sample it and distant enough to ask naive questions. The sceptic picks the hash samples and writes two of the five drill questions.
What if the review finds something serious?
Then it has done the best work a review can do. Convert it to a task with an owner and a date, fix it through the proper paths — imports unchanged, notes in the thread — and let the fixed record show the correction honestly.
Should the auditor see our review notes?
Notes filed in the room are part of its record, readable under the auditor's grant. Most teams find that an asset: a dated series of self-examinations is strong evidence of a functioning control environment.
How long should a review take?
About ninety minutes for one room, including the drill. If it regularly takes longer, the filing rhythm between reviews is where the time went missing.
Keep reading
All AI Audit Exchange guides · Auditor directory · About AI Audit Exchange · Security · Pricing · AI Audit Exchange home · Previous: Who Needs a California AI Audit? Working Out Your Position · Next: Evidence Room Security: How Access and Accounts Are Controlled
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