Boards don't need to become technologists to oversee AI, but they do need to ask five questions consistently: Where is AI used and who approved it? How do we classify AI by risk, and who owns that? How do we manage AI we buy from vendors? What happens tomorrow if an AI system causes harm? And who is the senior leader accountable for AI, with what authority? Vague answers to any of them are the finding.
Free AI leadership role brief
If the answer to question five is unclear, start there. After a 30-minute call we send a one-page brief on the role: scope, reporting line, a pay range from public filings, and where the candidates are.
Get my role brief →Why do boards need structured AI oversight?
AI has moved into hiring, pricing, customer service, product development and risk assessment faster than board oversight has followed. That gap creates regulatory, reputational and operational exposure, and rules such as the EU AI Act, whose high-risk obligations apply from December 2, 2027, make it concrete. Effective oversight comes from asking the right questions consistently, with enough structure to support decisions.
1. Where is AI deployed, and who authorized each use?
It sounds basic; it isn't. AI often arrives through individual teams, through vendor software with AI features nobody flagged, and through prototypes that slipped into production. A board that can't get a clear answer has a governance problem before any regulation applies. Every other obligation depends on knowing which systems you have.
2. How do we classify AI by risk, and who owns that?
A chatbot answering return-policy questions is not the same as a system that screens job candidates or evaluates loan applications. The board should know how management classifies AI systems by risk, what criteria it uses, and who can escalate or stop a deployment that crosses a threshold. If there is no formal classification, that is the most important finding of the conversation.
3. How are we managing third-party AI risk?
Most organizations buy AI rather than build it. Under the EU AI Act, the vendor that provides a system and the organization that deploys it each carry their own duties, and the deployer's duties don't pass to the vendor. Ask what due diligence is done on AI vendors, what the contracts say about data handling, model updates and incident response, and whether anyone in-house can test vendor claims about accuracy, bias and safety.
4. What would we do tomorrow if an AI system caused harm?
Treat it as a stress test. A recommendation engine that excludes a group, a hiring tool that disadvantages certain candidates, a credit model whose denials can't be explained: these are governance events that bring regulatory scrutiny, litigation and reputational damage. Ask who gets notified, how fast a system can be suspended, and what records would show the organization acted responsibly. Vague answers mean the organization isn't ready for the failure that matters most.
5. Who is accountable for AI governance, and do they have enough authority?
This ties the others together. Governance without a named owner is policy without enforcement. The board should know who owns AI governance at executive level, who they report to, whether they have budget and authority, and how their work is measured. Be skeptical of models where AI is shared informally among the CIO, CTO, CDO and General Counsel: in practice that often means nobody owns it. The CAIO role has arrived covers what the role looks like.
What should the board do with the answers?
Expect uncomfortable answers; that's the point. Oversight works when directors expose gaps before regulators, plaintiffs or the public do. Record the questions and answers in the minutes, set dates for follow-up, and consider whether AI needs a home in a committee. Why your board needs an AI committee and what boards get wrong about AI risk go further.
What are other boards disclosing about AI oversight?
In proxy statements filed with the SEC from January to October 2026, 600 of 4,667 (12.9%) say the board or a board committee oversees AI. Of the 319 that name a committee, 189 (59%) give the job to the audit committee, and only 6 name a committee after AI. 457 (9.8%) describe a director with AI experience or expertise. See our full table of 2026 proxy disclosures.
Frequently asked questions
What questions should a board ask about AI?
Five: where AI is deployed and who authorized it; how AI is classified by risk and who owns that; how vendor AI is managed; what happens if an AI system causes harm; and which senior leader is accountable for AI governance, with what authority.
Does the board need to understand how AI works?
Not in technical depth. Directors need enough literacy to ask these questions, judge the answers and recognize when management can't answer them. Many boards add depth through an AI-literate director, an advisor or a committee.
Is the vendor responsible if a purchased AI tool causes harm?
Not entirely. Under the EU AI Act, the vendor has provider duties and the organization using the tool has its own deployer duties, such as human oversight and monitoring, which can't be passed back to the vendor.
Written by Mariya with research and drafting help from AI tools. Every figure was checked against the sources below.
Sources
- Regulation (EU) 2024/1689 (EU AI Act), EUR-Lex.
- EU AI Act high-risk dates as amended by the 2026 Digital Omnibus on AI: December 2, 2027 (Annex III) and August 2, 2028 (Annex I).
- SEC EDGAR DEF 14A filings, January to October 2026, compiled in Talent Echo, Board AI oversight in 2026 proxy statements.