Back to Guides
Enterprise Software 15 min read

What a board needs from AI education, and what it does not

What a board needs from AI education, and what it does not

A board does not need to learn how to use AI. It needs to be able to tell whether management’s answers about AI are any good, and to stop leaking its own board papers while it works that out. Almost everything sold as AI education for directors teaches the skill the company’s staff need, and neither of the two a board does.

The ask usually arrives secondhand. A director says something at the end of a meeting, the chair agrees the board should get up to speed, and the task lands on a chief of staff who now has to buy something. What a search returns is a wall of priced programs: three days in a classroom, a 22-hour certificate, a half-day briefing from the firm that audits you. Some of it is worth buying. Most of it answers a question nobody at a 50-to-500-person company asked.

Two jobs wearing one name

“AI for board directors” covers two responsibilities that happen to share a room.

The first is oversight of the company’s AI: whether management has a governance structure, whether the risk sits inside the risk appetite the board already set, whether the investment case holds. That is duty of care applied to a new subject, not a new doctrine, and it is discharged by demanding reporting the board can interrogate.

The second is the board’s own use of AI: directors summarizing board books and benchmarking peers with consumer chat tools on personal accounts. That is happening now and has no owner.

Programs concentrate on the first, because the first is a curriculum and the second is an awkward conversation. The second is where the exposure is.

The four things a board actually needs

Strip the catalog language away and the content is short.

What it needs Why How long
Enough comprehension to read management’s AI claims So a confident deck does not pass unchallenged 30–40 minutes
A map of where AI already touches the company Most boards do not have one and assume someone does 20 minutes, mostly management’s homework
A felt sense of how these tools fail Secondhand knowledge of hallucination does not transfer; seeing one does 30 minutes, hands on keyboards
The data boundary for board material itself Because the board is a deployer of AI too 15 minutes and one written rule

The third line is the one people argue about. A director does not need tool proficiency. A director needs to have watched a current model produce a clean, well-formatted, entirely wrong answer about something the director actually knows. Thirty minutes of that changes how a board hears the phrase “the model found” for years, and no briefing deck does the same work.

Trade commentary agrees on the ceiling: directors need fluency to challenge management and judge trade-offs, not technical expertise. The argument is about the floor, and the floor sits lower than the price list implies.

The six things boards get sold

  1. A dedicated AI expert director. The Harvard Law School Forum on Corporate Governance worked through this in April 2026. Qualified candidates are scarce and the group dynamics cut the wrong way: “other directors may defer excessively to an AI expert,” and concentrating the knowledge in one seat “may also reduce other directors’ incentives to learn about AI.” A board of eight that outsources a subject to one member has created a point of deference, not competence.
  2. A lecture on how the models work. Training-data volumes and attention mechanisms have never changed a board decision. The decision-relevant facts are behavioral: these systems are fluent regardless of accuracy, inconsistent across identical prompts, and unable to flag what they do not know.
  3. The staff workshop at a higher price. The most common mis-purchase. Applied drafting and verification practice makes a finance team faster and is the wrong content for a board, whose working material is judgment about other people’s work. We lay out why the curriculum changes by function in our role-by-role view of enablement.
  4. Writing the AI policy. Management writes it. The board reads it, tests it against two or three uncomfortable scenarios, and asks who enforces it. A board that drafts the policy has become management and has nobody left to hold accountable.
  5. A per-director certificate as the opening move. Certificates give depth to the one or two directors who will carry the topic on a committee. For a whole board they cost a lot and do not produce the shared vocabulary you wanted, because modules completed privately over six months produce six private understandings.
  6. A vendor demo. A session run by a company selling a platform teaches the platform. Directors leave able to describe one product and no better at judging a proposal.

The urgent half is the board’s own tool use

Here is the pair of numbers that should restructure the agenda, both from Diligent Institute’s June 2026 Director Confidence Index, a survey of 104 US public company directors.

Eighty-two percent of those directors had used generative AI in their board work in the previous six months, up from 66% nine months earlier. Fifty-four percent said their company has no guidance on director AI use, only 6% reported a policy specific to the board, and around 30% had used it to summarize board books.

Then the two findings that cannot both be comfortable: 49% had heard of board members using publicly available consumer AI tools for board work, while 96% said they knew of no instance of board materials being entered into a tool outside an approved environment.

Read those together. Half know colleagues are working on personal accounts, and nearly all are confident nothing sensitive has crossed over. Both beliefs can be sincere, because a director has no mechanism to know: no log, no admin console, no retention setting a board can inspect. The confidence is structural, not evidential.

A separate Deloitte survey of governance professionals, reported in July 2026, found 51% of boards lack rules and guidance on AI use and 8% use company-provided standardized tools for committee processes. Deloitte called board-level AI use “comparatively new, uneven and still maturing.”

So the first deliverable of board AI education is not a strategy discussion. It is one page: which tool, in whose account, with what retention, and what never goes near a model. The Australian Institute of Company Directors publishes a usable red-flag version of this, and two of the flags it raises are board packs analyzed in public tools and directors relying on a generated summary instead of reading the papers. The second is a governance failure dressed as a productivity gain. It is the same artifact a legal function writes for itself in a working session, and the board has to write it rather than receive it.

Ninety minutes, then a standing item

The unit that fits is 90 to 120 minutes once, inside or adjacent to a regular meeting, followed by a standing agenda item of 15 to 20 minutes.

A defensible run of show:

  • 15 minutes — management presents the map: every place AI touches the company, who approved it, what data it reaches. Building the map is most of the value, which is why it is management’s homework.
  • 30 minutes — hands on keyboards. Directors put a question they know the answer to into a current model, then check a figure it produced.
  • 25 minutes — the questions block, worked against the company’s real situation, not in the abstract.
  • 20 minutes — the board’s own boundary, written and minuted before anyone leaves.
  • 15 minutes — the cadence: what reporting the board wants, from whom, how often, and which committee holds it.

Run it after the executive team has done its own session, because a board ahead of management generates recommendations nobody can act on. The questions a leadership team brings are a close preview of what the board will ask, which we catalog in our account of the questions leadership teams actually raise.

The cadence matters more than the session. Protiviti and BoardProspects surveyed 772 board members and executives in the fourth quarter of 2025 and found only 26% of boards discuss AI at every meeting. Among organizations reporting high returns on AI, 63% had it on every agenda, against 13% of low-return organizations. That is correlation, as the press release says. It still suggests a one-time briefing with no standing item does nothing.

What the market charges

Verified against each provider’s public page in October 2026.

Option What it is Price
NACD and Carnegie Mellon Heinz certificate 10 asynchronous modules, 22 hours, six-month window, exam, 22 NACD credits $3,995 corporate members, $4,495 individual members, $4,995 nonmembers
Columbia Business School Executive Education Three days in person, committee-by-committee curriculum $8,400
INSEAD board AI program Assessing an organization’s AI capability from a board seat Not published
Private session for the whole board 90 to 120 minutes on the company’s own material, plus a cadence Facilitated workshops run roughly $2,000 to $15,000

Catalogs change quarterly, so confirm before budgeting. The arithmetic is per-seat against per-room: seven directors through a certificate at nonmember rates is a five-figure spend producing seven private educations, while one session produces a shared standard, a boundary, and a cadence, and leaves budget for the committee chair who should go deeper. Our comparison of executive and leadership programs runs the same math.

Eight questions a director should be able to ask

Board AI education produces this list, and the nerve to use it.

  1. Where does AI touch our company today, who approved each instance, and what data does each one reach?
  2. Who owns AI governance by name, and how many hours a week do they have for it?
  3. What is the one decision we have delegated to a model, and what happens when it is wrong?
  4. Which of our vendors is running a model over our customer data, and what does that contract say about training?
  5. What did we spend, and what did it return, measured how? A bad answer cites license seats and weekly active users.
  6. What are we spending on retraining and compliance that is not in the technology line?
  7. Which competitor move would change our plan, and would we notice in time?
  8. What is the board’s own rule for AI, and has anyone broken it?

Question five is where boards get the least from their time. Usage dashboards are not returns, the subject of our longer argument about why licenses and capability are not the same purchase. A board that accepts a seat count as a return has taught management that a seat count is acceptable.

Duty of care, and the one statutory obligation

The reliance rule is more generous than directors fear. Under Section 141(e) of the Delaware General Corporation Law, a director who relies in good faith on officers, employees, or experts selected with reasonable care is “fully protected” on matters within their competence. Baseline literacy plus regular reporting is a defensible posture. The catch at this company size is that there may be no CTO to rely on, which raises the floor on what the board itself must understand.

The statutory literacy duty is narrower than vendors imply. Article 4 of the EU AI Act requires providers and deployers to “take measures to support the development of AI literacy” among staff, and has applied since 2 February 2025. It does not require guaranteeing any specific level for any individual, and it bites only where there is EU exposure. Treat it as a reason to document what you did, not as a reason to buy a certificate.

Disclosure practice moves faster than either. EY’s Center for Board Matters, reviewing Fortune 100 filings through July 2025, found 48% of companies citing AI risk as part of board oversight, up from 16% a year earlier, and AI expertise in 44% of director skills matrices, up from 26%. Private boards have no proxy statement to fill, but their investors read public ones.

How to tell in two quarters whether it worked

Three checks, all readable from the minutes:

  • The boundary exists and someone has tested it. One page, written by the board, with a named owner.
  • AI appears in the minutes without being an agenda item. It shows up inside a budget discussion or a vendor approval, not only in its own slot.
  • Someone pushed back. A director challenged an AI claim. If that has not happened in two quarters, the session bought agreement, not competence.

Frequently asked questions

Does every director need AI training, or only the committee that owns it?

Every director needs the baseline; one or two need depth. The Harvard Law Forum’s position is that all directors should possess at least baseline AI literacy. Depth belongs with whoever chairs the committee holding AI oversight, and that seat is the one worth a certificate.

Should the board be educated before or after the executive team?

After, by a few weeks. A board ahead of management produces direction nobody can execute. Running leadership first also tells you which questions the board will raise.

Do we need to hire a director with AI expertise?

Probably not. Specialists are scarce, and a designated expert tends to attract deference and reduce everyone else’s incentive to learn. Get the baseline across the board and buy expertise as advice.

What should the board’s own AI rule say?

Four things: which tool is approved, whose account it runs in, what happens to the data, and what never goes in. One page, written by the board, because a rule handed to directors gets read once.

Is it acceptable for a director to read an AI summary instead of the board papers?

No, and it is one of the clearest red flags in the governance guidance. A summary is a reasonable-sounding compression by a system that cannot tell which omission matters. Use one to orient, never instead of reading.

How does this differ from AI training for employees generally?

The content differs because the work differs. Staff training builds applied practice on the team’s own documents and daily tasks. Board education builds judgment about other people’s work, plus one hands-on block to calibrate it.

What does board AI education cost?

Per-seat programs run from roughly $4,000 for an asynchronous certificate to $8,400 for three days in person at a top business school. A private session for the whole board sits in the $2,000 to $15,000 range typical for workshops.

Our board is investors and advisors rather than independent directors. Does this still apply?

The oversight half scales down and the self-governance half does not. An investor director reviewing your board deck on a personal account is the same confidentiality exposure as a public-company director doing it, with less infrastructure around them.

Key takeaways

  • Board AI education covers two jobs: overseeing the company’s AI, and governing the board’s own use of it. Programs teach the first. The second has no owner.
  • The requirement is short: comprehension enough to read management’s claims, a map of where AI already touches the company, 30 minutes of watching a model fail, and a written data boundary for board material.
  • In Diligent’s June 2026 survey of 104 public-company directors, 82% had used generative AI for board work, 54% had no guidance, 49% had heard of colleagues using consumer tools, and 96% knew of no instance of material leaving an approved environment. Nobody can verify the last one.
  • Skip the dedicated AI expert director, the architecture lecture, the staff workshop at a markup, drafting the policy, and a certificate for everyone before the board has a shared vocabulary.
  • Delaware’s Section 141(e) protects good-faith reliance on management, and EU AI Act Article 4 asks for measures toward literacy, not a guaranteed level. The floor rises when there is no CTO to rely on.

If your board has asked to get up to speed on AI and you are the one who has to buy something, book a free AI readiness call. We will tell you what the room needs, what to skip, and what it should cost.

Last Updated: Nov 15, 2026

AW

Arthur Wandzel

SFAI Labs helps companies build AI-powered products that work. We focus on practical solutions, not hype.

Make your team fluent in AI — then automate what proves out

  • Hands-on training applied to your own documents and workflows
  • Often reimbursable through state small-business training programs
  • Built for 5–50 person firms with no IT department

Related articles