When the court ordered Reaves Law Firm to produce evidence of its AI verification process, the firm couldn't. No training records. No review logs. No documented oversight. Just an internal email with a subject line about protocols and a claim that processes had been "restructured." The firm was sanctioned under Rule 11, not because it used AI, but because it couldn't prove it had verified what the AI produced.
You don't want to be in that position. This template gives you a framework to document AI oversight in a way that produces evidence, not just policy statements.
Purpose of the Template
This verification log template is for any process where your organization uses AI to generate content, recommendations, or analysis that feeds into consequential decisions. It creates contemporaneous records that answer four questions a regulator, auditor, or court will ask:
- What did the AI system produce?
- Who reviewed it and against what standard?
- What judgment did they exercise?
- What conclusion did they reach?
Use this when AI output influences decisions with legal, financial, or reputational consequences: regulatory filings, contract terms, compliance assessments, risk ratings, audit conclusions, or customer communications.
Prerequisites
Before you implement this template, ensure you have:
A defined scope. Identify which AI-assisted processes require human verification. Not every AI interaction needs this level of documentation, but any output that could expose your organization to liability does.
Named owners. Assign a specific person to each verification task. "The compliance team" isn't accountable. Sarah Chen, Compliance Analyst, is.
Review standards. Document what constitutes adequate verification. For legal citations, confirm the case exists and supports the proposition. For risk assessments, validate that the AI's data sources are current and that its weighting aligns with your risk appetite.
A storage system. These records need to be retrievable. A shared drive folder with consistent naming conventions works. A GRC platform with audit trails is better.
The Template
Create a verification log for each AI-assisted decision. Save it with a clear identifier: AI_Verification_[Process]_[Date]_[Owner].docx
AI Output Verification Log
Process: [e.g., SEC Form 10-Q risk factor drafting]
Date of AI Generation: [YYYY-MM-DD]
Date of Verification: [YYYY-MM-DD]
Verified By: [Full name and title]
AI System Used: [Name and version, e.g., GPT-4 via Azure OpenAI]
AI Output Summary:
[Paste or describe what the AI generated. If it's a document, include the file path and version.]
Verification Standard Applied:
[What criteria did you use to evaluate the output? Reference your internal policy section if applicable.]
Verification Steps Performed:
- [Specific action taken, e.g., "Cross-referenced all cited regulations against current CFR text"]
- [Specific action taken, e.g., "Validated financial figures against Q2 management reports"]
- [Continue for each verification step]
Issues Identified:
[List any errors, hallucinations, or unsupported claims found in the AI output. If none, state "None identified."]
Corrections Made:
[Describe what you changed and why.]
Final Judgment:
[Your conclusion: "Output approved as written" / "Output approved with modifications documented above" / "Output rejected; alternative approach required"]
Approval Signature: [Digital signature or typed name]
Supervisor Review (if required): [Name, date, signature]
Customizing the Template
For regulatory filings: Add fields for specific regulatory citations verified and a checklist of disclosure requirements confirmed. Include cross-references to your obligations library entries.
For risk assessments: Add fields documenting data source validation, assumption testing, and comparison to previous risk ratings. Include a section for professional skepticism notes where you document why you questioned the AI's conclusions.
For contract review: Add fields for specific clauses reviewed, deviations from standard language identified, and legal precedent confirmed. Reference your contract playbook standards.
For audit work: Add fields mapping AI findings to control objectives, evidence reviewed to support AI conclusions, and sampling methodology validated. This becomes part of your audit fieldwork documentation.
For multi-stage processes: Create a master log that references individual verification logs at each stage. This shows the cumulative oversight path from AI generation to final decision.
Validation Steps
Before you rely on this system under pressure, test it:
1. Conduct a mock audit. Ask your internal audit team or outside counsel to request evidence of AI verification for a recent decision. Can you produce the complete record within 24 hours? If not, your storage and retrieval process needs work.
2. Run a failure scenario. Deliberately introduce an error into an AI output and see if your verification process catches it. Document the catch in your log. This proves your process works and creates a record of your quality controls functioning as designed.
3. Check for completeness. Pull five verification logs at random and confirm every required field is populated. Incomplete logs are worse than no logs because they suggest a process that exists on paper but isn't followed in practice.
4. Verify ownership. Confirm that every person listed as a verifier in your logs from the past 90 days is still employed and still assigned to that function. If you can't produce the verifier when asked, you have a gap.
5. Test your escalation trigger. If a verification log shows repeated errors from the same AI system, does that trigger a review of whether the system should continue to be used? Document that decision path.
The Reaves case turned on a simple question: What steps did you take to verify? The firm couldn't answer. Your verification logs should make that question trivial to answer. If you're ever served with a show-cause order, a regulatory inquiry, or an audit request, you produce the log. The evidence speaks for itself.
That's the difference between having an AI policy and having AI governance that can withstand scrutiny.
