Why Directors Need the AI Disclosure Evidence List Before They Approve the 10-K
Should your directors rely exclusively on management for the AI language in your next 10-K, when AI is the one risk with a growing case docket behind it?
Reliance on management is a defense right up to the point where no one can produce the evidence. Your audit committee approves the AI language before the annual report is published. Plaintiffs quote it afterward.
The Situation
Total securities class action filings rose 30% in the first half of 2026, and the AI-related cases within that nearly match the total number of AI filings in all of 2025 - putting the year on pace to almost double. Cornerstone Research and the Stanford Law School Securities Class Action Clearinghouse measured these cases by Disclosure Dollar Loss, defined as the market value that evaporates on the day a company discloses the problem, and those AI cases produced $385 billion of the $529 billion total market value loss, or 73% of it. Those losses trace to statements executives made about what their AI was doing, and to the absence of anyone inside the company who could produce evidence for them.
The Exposure
The ZoomInfo complaint filed in the Western District of Washington last June alleges the company told investors its go-to-market AI was driving stronger daily engagement, while in fact customers were rejecting the AI products and retention was declining, and the stock fell more than 33% when guidance came down. A claim that AI is driving engagement is verified with retention, adoption, and usage data rather than the model documentation that AI governance programs usually focus on. The SEC set the standard in its 2025 Presto Automation order, finding that the company "had no established process for drafting, reviewing, or approving" its filings and that "no one at Presto was formally responsible" for their accuracy. The Conference Board found 83% of S&P 500 companies now disclose AI as a material risk, up from 12% in 2023, which means the volume of AI statements in filed documents has grown faster than the process for standing behind them.
The Judgment Call
Reliance on management is how boards work, and singling out one topic for detailed director-level attention is a departure from how the audit committees generally review filing materials. But reasonableness is what makes reliance a defense, and it’s measured against what the director knew at the time. AI-related securities filings are on pace to nearly double last year's total, and they now carry an SEC enforcement template and a loss share running more than 5X their filing share, which puts AI disclosures in a different category than the others. Directors should require the evidence list for AI statements in the pre-read, not at the table, covering every AI claim in the annual report and 10-K, and each with the name of the person who can produce what stands behind it. For directors of private and sponsor-owned companies, identical claims appear in lender covenant representations, purchase agreement representations, and the D&O renewal application signed every year.
Risk: Management will read a director-level evidence request as a signal that the board doesn't trust the disclosure process, and your CFO and General Counsel will point out that it arrives during the tightest window in the reporting calendar.
Benefit: A board that received the evidence list before it approved the language holds a contemporaneous record of the inquiry, which is the difference between reliance a court finds reasonable and reliance a plaintiff characterizes as a rubber stamp.
This Week’s Action
What to do: Ask your audit committee chair to add one item to the pre-read for the meeting where the 10-K is approved: a schedule listing every AI-related statement in the draft, the named individual accountable for each one, and the date of the evidence that supports it. Request it a week ahead, not at the meeting.
Who to involve: Your audit committee chair to make the request, your CFO, who certifies disclosure controls, your IR officer, who holds the transcripts where most AI statements originate, and your CAE to compile the schedule independently.
What outcome to achieve: The schedule in your director’s hands before the meeting, and minutes reflecting that the committee asked for it and received it. The schedule protects the company. The minutes protect the directors who approved it.
Time required: 15 minutes to frame the request with your chair, 45 minutes to review the schedule before the meeting.
Artifact
Five Questions Before You Approve the AI Language
Run these against each AI-related statement in the draft filing, one statement at a time.
1. Named Owner - Does a single named individual, not a function or committee, hold accountability for the accuracy of this statement?
→ YES: Record the name in the minutes next to the statement.
→ NO: This is the Presto profile; don't approve the statement until a name is attached to it.
2. Independent Behavioral Evidence - For any claim about the AI, can the company produce a test of the model's actual behavior run by someone other than the team that built/deployed it, supporting the claim as written?
→ YES: Confirm the test is repeatable and the results are retained where counsel can retrieve them on 48 hours' notice.
→ NO: The claim rests on the assurance of the team whose work it describes. Independent behavioral assurance testing is a control you can commission, and it should exist before the next 10-K issuance.
3. Commercial Evidence - For any claim that the AI is producing results, whether adoption, engagement, efficiency, or retention, can Finance or Operations produce the underlying data on the same basis and for the same period the statement describes?
→ YES: Confirm the figures in the filing reconcile to the data set Finance/Ops produced.
→ NO: Strike the claim, or qualify it down to what the data actually supports.
4. Contradiction Check - Has anyone asked management directly whether any internal reporting, customer retention data, or pipeline metric points in the opposite direction from the public AI claim?
→ YES: Document the question and the answer in the minutes. That record is the defense.
→ NO: Ask it at the meeting. This is the question the ZoomInfo complaint was built on.
5. Cross-Document Consistency - Do the AI representations in your D&O renewal application, lender covenant certificates, and any outstanding purchase agreement reps say the same thing as the filing?
→ YES: Confirm who reconciles them annually, and when.
→ NO: You have inconsistent representations across documents signed by different officers at different times. Get it resolved before the filing goes out, not after a carrier denies your claim.
If your General Counsel hasn't reconciled the AI language across those documents, that's the first task to initiate.
Working through a specific AI decision right now? Reply to this email and tell me what it is. I read every response and answer directly.
When the stakes exceed your internal capacity:
AI Exposure Diagnostic: A 2-hour strategic evaluation for risk, compliance, and legal leaders to identify your highest-priority governance gaps and deliver a 90-day remediation roadmap.
12-Week Governance Sprint: Translate regulatory requirements into audit-ready policies, control frameworks, and accountability structures.
Fractional Chief AI Officer: Embedded ownership of governance, intake, and board reporting before the function is formalized, with the eventual role scoped and the search supported when it's time to hire.
Reply with "Diagnostic," "Sprint," or "Fractional" to schedule a conversation for next month.
Chris Cook writes Judgment Call weekly for compliance and risk officers navigating AI governance.
Former IBM Vice President and Deputy Chief Auditor. Published in the AI Journal, speaker at Yale.
Chris Cook
Managing Partner & Founder
Blackbox Zero
Forwarded this by a colleague? Subscribe to Judgment Call
