The 2026 OnBoard Board Effectiveness Survey · Sixth annual report
And 60% of them sit on boards with no formal AI policy. The open question is no longer whether directors are using AI on the most sensitive material a fiduciary handles — it is whether the board has written down how, and built a secure place for them to do it.
The 6% of boards that have written and enforced a policy rate themselves 32 points more effective than the 38% that have nothing. That gap, across every metric in this study, is the story.
— Five things to take from this report
Sixth annual report · 531 governance professionals · survey closed June 1, 2026. Year-over-year figures compare to the 2025 report (n = 549). Every figure is computed against the raw response file.
00 — What changed since last year
In 2025, AI took a firm foothold in the boardroom with 69% of directors using AI to support board work. By the time the 2026 survey closed nine months later, 89% of all respondents — and 92% of board directors specifically — were using AI for board work. Critical mass in adoption stretched across roles, sectors, and geographies.
What did not arrive alongside adoption was the governance scaffolding. 63% of boards still have no formal AI use policy. Only 6% have an enforced policy requiring signatures and reviews. It should come as no surprise to see that confidence in board security dropped 15 points. The most-cited negative impact of AI on board effectiveness is data privacy — a direct symptom of the gap between use and oversight. That is the story this report tells, and the title is the data’s own diagnosis: boards adopted AI faster than they governed its use. Every leader, every director reading this report is most likely sitting on a board that has crossed the adoption line and not yet instituted an enforced policy. This research shows a counterintuitive reason they should.
01 — The Shadow AI Boardroom
8% of directors put board work through DeepSeek — a Chinese-built open-source model. The dominant workflow runs these tools against the materials before the meeting: summarizing the board book, anticipating questions, researching the topic on the table.
Summarizing a board book is the same as uploading a board book — strategic plans, M&A diligence, compensation reviews, legal opinions, audit findings. By default, consumer accounts train on uploads unless a director is on a paid tier with training disabled.
The first row is disquieting. Summarizing a board book is the same as uploading a board book: they are uploading strategic plans, M&A diligence, executive compensation reviews, legal opinions, audit findings, vendor risk assessments, and security incident updates. This is the most sensitive material a director sees as part of their fiduciary duty. By default, consumer-LLM accounts train on uploads and conversations. Those documents are entering the LLM’s corpus unless the director uses a paid personal account and has manually disabled the training setting, or is using an enterprise tier account.
The phrase our research team kept coming back to in the analysis was the Shadow AI Boardroom — directors sending sensitive board materials through tools their boards have not approved, do not control, and in many cases do not know are being used. Across this dataset, that behavior is the default.
Two patterns make the exposure compound. The first is that directors are using several different AI models in their work: 60% of board professionals run two or more AI tools (up from 39% in 2025). The second is who’s in the cohort. Those with the deepest fiduciary exposure — directors and CEOs — are also the heaviest, broadest users of AI. At some level, these individuals recognize the need for security and are the most likely (39% each) to rate their interest in a secure AI solution at 8 or higher.
Respondents named the consequence. When asked where AI has negatively impacted board effectiveness, data privacy, confidentiality, and data usage concerns took the top spot at 38% — ahead of every other answer. In second place, respondents reported their board effectiveness was negatively impacted by revealing limited AI knowledge and expertise among board members (27%). Board members are looking around the room, recognizing when a peer is not keeping pace, and it is impacting their view of the entire board.
Director takeaway. The probability that multiple members of your board put your board book through a consumer AI tool in the last ninety days is north of 90%. The open question is whether your board has written down what it expects of itself in a formal policy.
The cohort with the deepest fiduciary exposure — directors and CEOs — posts the highest intent. The people who recognize the need have not been given the system.
02 — The Governance Gap
63% of boards have nothing formal in place. 6% have a policy with enforcement behind it. The middle is wide and active — a quarter of boards are drafting their policy right now.
When our research team looked at whether adopting an AI policy moved other effectiveness metrics, the pattern was striking. Effectiveness, security confidence, collaboration, perceived AI impact, and investment intent all climbed in step with policy maturity.
Each respondent was asked, “Compared to 12 months ago, has the success of your board improved or declined in the following areas?” and given a five-point scale to evaluate their effectiveness, collaboration, and confidence in security.
The lift is consistent across all five stages of policy maturity. If policy were a stand-in for “well-funded, sophisticated boards,” the curve would jump only at the top. Instead, the number climbs at every step — which is the first signal that something other than board sophistication is doing the work.
The biggest single-step lift is the move from Developing to Basic guidelines — +11 points. That is the moment a board’s AI posture stops being a conversation and becomes a written artifact. The act of writing the policy down appears to do meaningful work in the perception of a board’s effectiveness, security, and collaboration.
Shared formal policy plateaus slightly below Basic guidelines on effectiveness. Boards that socialize a policy without enforcing it are sitting in a trap — the artifact has been delivered, the behavior has not necessarily followed. Enforcement is where the curve resumes climbing.
Security confidence has a different shape from effectiveness. Tiers 2–4 are essentially flat (53% to 57%). Then it jumps 24 points at enforcement. Writing a policy improves perceived effectiveness on day one. Enforcing it is what makes a board feel safer.
The directors at the top of the curve are running more AI, in more places, while governing the work they do with each one. And as policy matures, AI moves into the highest-judgment work. The share of directors using AI to anticipate board questions and prepare responses grows from 12% in no-policy boards to 67% in enforced-policy boards — a 5.6× lift. Summarizing board books grows from 50% to 80%. Governance review grows from 26% to 47%. Policy turns out to be the gate that brings the higher-stakes work into scope.
It’s worth noting that boards with an enforced policy are more likely to be corporate, more likely to be larger in size, or resourced enough to write and enforce a policy in the first place.
| Policy tier | Effective | Secure | Collab. | Positive AI | Invest 8–10 |
|---|---|---|---|---|---|
| No policy | 55% | 28% | 56% | 51% | 20% |
| Developing | 70% | 53% | 68% | 73% | 33% |
| Basic guidelines | 81% | 53% | 74% | 79% | 43% |
| Shared formal | 79% | 57% | 73% | 79% | 35% |
| Enforced | 88% | 81% | 81% | 78% | 78% |
As policy matures, AI moves into the highest-judgment work: anticipating board questions grows from 12% to 67% — a 5.6× lift. Policy is the gate that brings high-stakes work into scope.
03 — Directors Lead Adoption and Carry the Most Exposure
When directors use AI, here’s what they are doing: 54% are reviewing board books and reports, 40% are researching topics and analyzing data, and 34% are drafting and editing presentations.
The data is clear: directors are using AI to prepare to perform their fiduciary duties.
ChatGPT and CoPilot show up in roughly 70–85% of director responses on every use case. What tools they use as the work gets harder is what has your general counsel concerned.
When directors anticipate hard questions, they typically reach for ChatGPT plus CoPilot plus Claude. For governance and regulatory reading, DeepSeek over-indexes by a wide margin.
That’s where the DeepSeek footprint gets uncomfortable. 31% of directors who use AI to review governance and regulatory content are running it on DeepSeek — a Chinese open-source model, applied to US, UK, and Canadian regulatory material on behalf of fiduciaries, with no board policy authorizing it and no IT team aware.
04 — Confidence in board security dropped 15 points
The 2026 figure sits below even the 2024 baseline of 54%. The drop is the second-steepest move anywhere in the survey.
Pair the drop in security confidence with the top negative-impact answer — data privacy, confidentiality, and data usage concerns at 38% — and the cause is not hard to find. Directors are watching the AI tool list grow, fearing their colleagues are uploading board books into ChatGPT, and recognizing that a crisis in information security is brewing.
The second negative impact pours salt on the wound. 27% of respondents say AI has revealed limited AI knowledge or expertise among board members.
This is the director looking around the room and recognizing that some of their peers do not understand the tools well enough to spot a bad answer, do not know what a hallucination looks like, do not realize that the model agreed with them because models agree by default. The bar for showing up to a board meeting has been raised, very concretely, by the directors at the same table who walk in readied by their favorite LLMs. Everyone else in the room can feel it.
The third negative impact — exposed gaps in AI governance policies (21%) — ties a bow around this issue. Directors know the gap is there. They named it. They have not been given the tools to close it.
Boards are collecting AI’s productivity dividend — efficiency in preparing materials (50%), more actionable minutes (43%), faster access to historical decisions (33%). They are waiting for governance to catch up.
Director takeaway. A 15-point drop in security confidence is the room telling itself the obvious. Without a written, enforced policy and a place to run board AI work that respects the permissions the board already operates under, this issue will compound.
05 — Where AI is helping — and where it has not landed
When asked where AI has positively impacted board effectiveness, respondents lead with the operational wins.
When asked, 50% of directors point to increased efficiency in preparing board materials. 43% cite more concise and actionable meeting minutes. 33% report faster access to relevant information and historical decisions.
What is missing from the top of that list is the rest of the story. Improved strategic scenario planning sits at 14%. Stronger risk identification and governance oversight at 7%. Increased engagement and participation during meetings, also at 7%.
The places where AI could most meaningfully change a board — strategy, risk, and engagement in the room — are not where boards are using it yet. That is the gap policy maturity closes: when boards climb the policy ladder, the judgment-grade use cases are what comes into scope.
05 — The Persistent People Problem Has Not Moved
87% of respondents report at least one ineffective member of their board.
We ask respondents each year what percentage of their current members effectively contribute to the board’s key goals and objectives. Every year, roughly the same answer: 87% of respondents report at least one ineffective member on their board. Most boards have ten or fewer members, so “at least 10% ineffective” reduces to “at least one person isn’t contributing.” The weighted-average estimated ineffectiveness sits at 37% in 2026 and 36% in 2025. Inside a sample of 531, that is noise.
The qualitative texture matches. The top opportunities-to-improve in 2026 are the same ones boards named in 2025: underperforming members, limited communication between meetings, unclear expectations, inconsistent preparedness. Two opportunities did rise materially — time constraints in meeting preparation (+5 points) and minimal accountability or follow-through (+4 points). The work is getting harder, even with AI in the picture.
AI changed what a prepared director looks like once they arrive at the meeting. The director who used to skim the board book on Sunday night is now sitting across the table from one who summarized it on Friday, asked their model what the board ought to push back on, and walked in with discussion points prepared. Their peers notice. 16.5% of boards named ineffective or slow adoption of AI as an opportunity to improve.
That observation stands as the strongest argument against the “AI board member” framing now appearing in the market. A set of vendors are selling AI assistants designed to act as board participants — or at least claiming that’s the future they envision. The data has nothing kind to say about that direction. The problem the data describes lives inside the seats already at the table — directors who could be operating at the full capacity their role deserves and currently are not. The work is to amplify every director already on the board, supported by tooling that respects the rules of the room.
Director takeaway. The composition of your board roster is a search-and-nominations problem. The preparation of the directors on it is the problem AI can actually move. Hold the distinction every time you read a vendor pitch.
07 — A framework
Across every metric we measured, boards with an enforced AI policy outperform boards with no policy. Effectiveness: 88% vs 55%. Collaboration: 81% vs 56%. Confidence in board security: 81% vs 28% — nearly three times higher.
What changes with policy is how directors use AI. Without a policy, only 12% use AI to anticipate board questions before a meeting. With an enforced policy, 67% do — more than five times higher. Summarizing the board book follows the same pattern: 50% to 80%.
Investment intent, though, doesn’t climb in a straight line. Boards with enforced policies are the most willing to invest in secure AI (78%). Next come boards still drafting their first set of guidelines (43%) — not boards that have shared a finished policy (35%). Boards drafting are actively shopping. Boards that have shared a policy without enforcing it have moved on, mentally — and they are the cohort most likely to discover, a year from now, that the policy was theater.
Each step is achievable. Start the conversation. Write the draft. Share the document. Put enforcement behind it. The next section breaks the work into four moves.
08 — What boards should do next
This document determines what a fiduciary may do with confidential board material. It belongs alongside D&O coverage and audit oversight. The single largest effectiveness lift in the dataset happens the moment intent becomes artifact.
92% of directors use AI; 60% of boards have no record of which tools. A short anonymous survey surfaces the actual perimeter the policy needs to govern.
Summarizing board books, anticipating questions, reviewing regulatory material — these require AI operating inside the same security envelope the materials already live in, drawing on the governance record, never training on board material.
The second-most-cited negative impact was limited AI knowledge among members. A consumer model on personal material teaches the wrong reflexes; the same model on the board's own record, inside a governed envelope, teaches judgment without putting material at risk.
09 — How OnBoard AI becomes a strategic partner to your board
Directors have already adopted AI for board work, faster than their boards have governed it. OnBoard AI is built against four rules this survey argues for.
Real minutes, real documents, real discussions. When a director asks OnBoard AI to summarize the board book — the use case 54% already run through consumer LLMs — it summarizes that board book, from inside the security envelope where it already lives. No hallucinations, no fabricated votes.
Committee membership, independent-director status, conflict screens — the information barriers already on your board apply inside OnBoard AI. A director asking about something they shouldn't see gets nothing, same as today.
Agenda suggestions draw on open items. Book summaries personalize to what each director flagged. Action items detected in the meeting flow into the next agenda. Nothing trains on your board's material.
Every meeting makes the next one smarter. A question raised in March surfaces in June's prep. An action item from Q2 appears on the Q3 agenda. Institutional memory no longer walks out the door when a director rotates off.
Methodology
