Liability

What Are a Director's Duties for AI Oversight?

Paula Fontana November 2026 8 min read

Directors are not expected to understand how an AI model works. They are expected to know where AI is material to the business, to have a system for hearing about it, and to be able to show that they acted on what they heard.

Key takeaways

  • A director's AI oversight duty in the U.S. is an application of the existing duty of oversight established in Caremark, not a new duty.
  • Courts have looked hardest at mission-critical risks. Where AI is central to products, safety or customers, the expectation of board-level reporting rises.
  • Where the company is incorporated shapes the legal standard directors are held to, so counsel should frame the analysis for your board.
  • The practical test is evidence: what the board asked, what it was told and what it decided, recorded at the time.

In the United States there is no AI-specific fiduciary duty. There is a general duty of oversight, and AI falls within it. In In re Caremark International Inc. Derivative Litigation (Del. Ch. 1996), the Delaware Court of Chancery held that directors must make a good-faith effort to ensure an adequate reporting system exists for the risks the company faces. Stone v. Ritter (Del. 2006) placed that obligation within the duty of loyalty.

The bar for liability is high. A plaintiff has to show that the board utterly failed to implement a reporting system, or consciously ignored red flags from one that existed. Most oversight claims are dismissed. But the cases that have survived share a pattern that is directly relevant to AI.

What have the courts said about mission-critical risk?

In Marchand v. Barnhill (Del. 2019), the Delaware Supreme Court allowed a claim against the board of Blue Bell Creameries to proceed after a listeria outbreak. Food safety was the company's central compliance risk, and the board had no committee, protocol or regular reporting on it. In In re Boeing Co. Derivative Litigation (Del. Ch. 2021), a similar claim over airplane safety survived because the board had no regular process for hearing about it. In 2023, the Court of Chancery confirmed in the McDonald's litigation that officers owe a duty of oversight within their areas of responsibility.

The thread is that when a risk is central to what the company does, the board is expected to have a dedicated way of hearing about it. For a growing number of companies, AI is becoming that kind of risk: it prices products, approves customers, screens candidates or runs operations. The more central AI becomes, the less a general technology update on the agenda will look like a reporting system.

Does where a company is incorporated change the standard?

Yes. At our Navigator launch with Dorsey & Whitney in Dallas, one point the room returned to was that a company's state of incorporation shapes what directors are legally accountable for. Delaware's oversight case law is the most developed. Nevada and Texas have codified different approaches to director protection, and Texas amended its business code in 2025. A board's AI oversight analysis should start with counsel confirming which standard applies, because the same governance gap can carry different exposure in different states.

What do disclosure rules add?

Oversight duties are enforced after the fact. Disclosure rules create obligations in advance. In March 2024 the SEC settled charges against two investment advisers for overstating their use of AI, the first enforcement actions against what is now called AI washing. Statements about AI capability in filings, earnings calls and marketing are now scrutinized for accuracy, and the board approves many of the documents they appear in.

Outside the U.S., the EU AI Act entered into force in August 2024 with obligations phasing in over several years. Its AI literacy requirement, which applies to providers and deployers of AI systems, took effect in February 2025. For companies with EU exposure, that makes AI competence an operating obligation, not only a board development topic. Our regulations guide covers the EU AI Act, SEC disclosure, the NIST AI Risk Management Framework and ISO/IEC 42001 in more detail.

What should a board be able to demonstrate?

Because oversight claims turn on what the board knew and did, the useful question is not whether the board understands AI. It is whether the board could show, from its own records, that it took AI seriously where AI is material. In the Vela Board Navigator℠, three of the 16 elements carry most of this weight:

  • Decision rights and escalation. Accountability for AI is clear across the board, its committees and management, and consequential AI decisions are escalated to the right level.
  • Board information and assurance. The board receives decision-useful evidence on AI adoption, outcomes, incidents and control effectiveness, not only progress updates.
  • Ethics, legal duties and defensibility. The board can explain how trade-offs on fairness, privacy, intellectual property and regulation were considered, and that explanation is documented.

A board that can answer those three well has done most of what the oversight cases ask. A board that cannot has a gap worth closing before an incident makes it visible.

Where should a board start?

Start by finding out where AI is material in your company, because that sets how much oversight is expected. Then look at whether your committee charters, agenda and board materials reflect it. The Board Snapshot takes five questions and produces a one-page brief your board can discuss at its next meeting. To see how these duties play out in a specific decision, Rehearse the Moment puts you in the seat for three minutes on a real scenario, such as The Eight-K Clock or The Footnote on Page 11.

This article is general information, not legal advice. Directors should consult counsel about the standard that applies to their company.

Common questions

Do directors need technical AI expertise?

No. The oversight duty asks directors to make a good-faith effort to ensure a reporting system exists and to respond to what it reports. Directors need enough understanding to ask informed questions and to know when to bring in outside expertise.

Is AI governance a board or a management responsibility?

Both. Management runs AI systems and the controls around them. The board oversees that work, sets the level of risk it will accept, and decides which AI decisions must be escalated to it.

Can documenting AI gaps increase liability?

Counsel should advise on your situation. The oversight cases have generally penalized boards that had no record of engagement, rather than boards that identified gaps and acted on them.

Ready to look at your board through the Navigator?

The Navigator is where every Vela engagement begins. Request access or a briefing to see what it surfaces for your board.