# AI deployment acceptance checklist

Publisher: Vouch
Reference: https://vouchapi.com/resources#deployment
Version: 2026-10-25

This is an operational review template, not a certification, legal opinion or guarantee of model accuracy.
Copy and adapt this checklist for your internal review. Keep the reference when sharing an unchanged copy.

## Workflow and ownership
- [ ] Name the workflow and the accountable business owner.
- [ ] Define permitted inputs, expected outputs and decisions the model cannot make.
- [ ] Name the human reviewer and escalation contact.
- [ ] Record acceptance criteria before testing.

## Data and permissions
- [ ] Confirm permission to process every input dataset.
- [ ] Record provider retention, training-use and access settings from current documentation.
- [ ] Test access with a user who must not see the restricted documents.
- [ ] Keep secrets and customer documents out of public examples.
- [ ] Separate permission to process data from permission to use it for research or training.

## Evidence and calculations
- [ ] Bind each material figure to its entity, metric, period, unit, document and page.
- [ ] Recalculate derived figures outside the language model.
- [ ] Check that the source supports the conclusion, not just the quoted number.
- [ ] Record unsupported claims as unresolved or remove them.
- [ ] Mark synthetic examples explicitly. Never present them as customer results.

## Failure and evaluation
- [ ] Test missing documents, conflicting sources and unsupported requests.
- [ ] Confirm the workflow abstains when evidence is unavailable.
- [ ] Keep evaluation document families separate from training material.
- [ ] Compare outputs using the same inputs and record model, settings, cost and review corrections.
- [ ] Verify the failure state reaches the user rather than silently substituting stale data.

## Release and records
- [ ] Have the owner approve the observed results and remaining limitations.
- [ ] Record model version, prompt revision, sources, output and approvals.
- [ ] Define rollback and incident handling.
- [ ] Confirm retention requirements with the responsible adviser or counsel.
- [ ] For applicable investment-adviser records, assess Advisers Act Rule 204-2 first.
- [ ] Do not describe a signature as proof of correctness or a legally mandated technology.

## Review record
Workflow:
Owner:
Reviewer:
Date tested:
Evidence location:
Open issues:
Decision: Hold / Approved with conditions / Approved
Conditions and next review:

## Product and regulatory references
- OpenAI Business: https://openai.com/business/chatgpt-pricing/
- OpenAI business data privacy: https://openai.com/business-data/
- Advisers Act Rule 204-2: https://www.ecfr.gov/current/title-17/chapter-II/part-275/section-275.204-2

Product settings and legal requirements can change. Consult current primary sources before approval.