Somewhere in the papers for your next governing body meeting there is a paragraph about AI. It says staff have had training, a policy is in place, and the school is keeping developments under review. Everyone will read it. Nobody will ask a question. The item will take four minutes, and the board will leave the room knowing precisely what it knew when it walked in.
That paragraph is not a report. It is reassurance with a date on it.
What should you report to governors about AI?
Report four things and then stop: what changed since the last meeting and when, what staff and students are now permitted to do that they were not permitted to do before, what went wrong or conspicuously did not go wrong in something you were watching, and the one decision you need from the board this term. If you need no decision, write the words "no decision is needed this term" and let the board hold you to that too.
Everything else is background, and boards are already drowning in background.
The test underneath all four is simple. After reading your paper, could a governor hold the school to something they could not have held it to before? If not, you have informed them. You have not let them govern. A board cannot oversee a position it was never given.
I use governors, trustees and board members interchangeably here, because the job is the same on both sides of the Atlantic: a group of volunteers, meeting a handful of times a year, expected to hold a school to account for something moving faster than their meeting cycle.
Why an update is not a report
An update tells a board what you did. A report tells a board what it now owns.
The gap between the two is where the risk on AI currently sits, and there is now evidence for how wide it has become. A study of 1,019 US K-12 education professionals and 1,029 parents, conducted by Morning Consult for IBM in July 2026 and published on 2 September 2026, found that 76 percent of middle school educators and 73 percent of high school educators use AI weekly, while only 20 percent have had extensive training in it. Only 24 percent of educators said US education is adapting well to advances in AI.
Read those numbers as a governor rather than as a school leader. Three quarters of the staff are using a technology weekly. One in five has been properly trained in it. And the people closest to it do not believe the system is keeping up. That is not a training gap. That is a live position the school is holding, and somebody has decided to hold it, even if nobody remembers deciding.
Across the leadership teams I work with, the AI paper that goes to the board is almost always written to close the item rather than to open it. It is written to survive the meeting. The trouble with surviving the meeting is that you survive it alone: nothing has been shared, so nothing has been carried.
The Governors' Test
The Governors' Test is three questions I ask of any AI report before it goes to a governing board: Does it tell them one thing they could not have assumed? Does it name the decision they are being asked to make, or the risk they are being asked to carry? And if any of it changed next term, would they hear it from us, or would they have to ask? A report that survives all three gives a board something to govern. A report that fails any of them is an update, and a board that only receives updates has not been governing. It has been kept informed.
This is my suggested way of reading a board paper, not a validated instrument, and it is deliberately unkind. Most AI items fail the first question. "Staff have received training" is something any governor would have assumed, because you told them last term that training was planned. It carries no information. A line of the shape "fourteen of our forty-one teaching staff used an AI tool to plan lessons last half term, and we know because we asked them, not because we measured it" carries information, including the admission about how you know. The figures there are illustrative; yours will be your own, and finding them is part of the job.
The third question is the one leaders skip. A board's real exposure is not what is in the paper. It is what will change between this meeting and the next one, in a term when a supplier can alter what a product does with your data without telling anyone. If the answer to "would they hear it from us" is no, you have quietly transferred a risk to people who do not know they are holding it.
The four lines a governing board should be given
Give the board four lines, in this order, and let the detail sit in an appendix nobody has to read.
One: what changed, with a date. Not "we have continued to develop our approach." A dated list. Three tools approved on 14 September. One withdrawn on 2 October after a supplier changed its terms. The teaching and learning policy amended at the October leadership meeting.
Two: what is now permitted that was not before, and who decided. The board needs the permissions in the words the staff were actually given, not a summary of them. If staff were told "you may use it for planning but never for anything that names a child," write that sentence in the paper. A governor should be able to repeat your rule to a parent in the corridor without reaching for the policy.
Three: what went wrong, or what did not go wrong that you were watching for. This is the line that separates an honest board paper from a brochure. If nothing went wrong, say what you were watching and why it mattered. A term with no incidents and no watch list is not a good term. It is an unobserved one.
Four: the decision you need, or the absence of one. One decision, named, with what happens if the board declines it. Boards handle a single real decision far better than five implied ones.
That is a page. It takes twenty minutes to write once you have the habit, and it changes the character of the meeting, because a board given four lines like these starts asking about line three.
What Good Looks Like
A well-governed school on AI is not the one with the longest policy. It is the one where a governor picked at random can answer three questions without notes: what is AI for in this school, what are staff not allowed to do with it, and who answers when it gets something wrong.
I publish a daily podcast and a newsletter that reaches more than 44,000 educators and leaders, which gives me a reasonably wide view of what leadership teams are worrying about in any given term. What arrives most often is not "what should our policy say." It is a quieter question: how would we know if this had already gone wrong? That is a governance question wearing an operational coat, and it belongs in a board paper rather than in a leadership team's inbox.
The strongest boards I see do one small thing differently. They ask for the same four lines every meeting, in the same order, so the shape of the paper never changes and only the content does. Drift becomes visible. A term where line three is empty twice running starts to look like what it is.
Three things boards are being told that are not quite true
Boards are currently being told three things that will not survive contact with the source documents, and a leader who corrects them earns more credibility than one who repeats them.
"The safeguarding rules now cover AI." In England, they barely do. I read the statutory guidance rather than the commentary about it, and Keeping Children Safe in Education 2026, which replaced the September 2025 version, contains a single reference to artificial intelligence: in the section on nudes and semi-nudes, noting that images "may also be digitally altered or wholly generated using artificial intelligence, including what are sometimes described as 'deepfakes' or 'deep nudes'." That is a real and serious addition. It is not a governance framework, it places no new AI duty on governing bodies or proprietors, and a board told otherwise will assume a floor that is not there. The floor is still yours to build.
"Everyone is banning it." On 2 September 2026, New York City, the largest school district in the United States, announced a one-year moratorium on student-facing generative AI from 2-K through eighth grade, affecting nearly 600,000 students, along with a ban on companion chatbots at every grade. Read past the headline and it is not a ban at all. Teachers may still use AI for planning and operational work. Assistive technology for students with disabilities is protected. Five high school pilots continue, capped at roughly 50,000 students. AI literacy modules run twice yearly for all high schoolers. That is not prohibition. It is a restriction with an end date, a protected list of exceptions, and a plan for what students learn instead. If your board wants to restrict something, that is the shape to copy.
"The research says AI makes students worse." The headlines from the launch of PISA 2025 on 8 September 2026 said something close to that. The OECD's own executive summary is more careful, and the care is the finding: "Students who do not use AI for specific schoolwork tasks outperform those who do, while weekly users and non-users of AI for general purposes perform similarly." It adds that "students tend to perform slightly better when frequent AI use is paired with opportunities to develop AI literacy skills." These are associations, not causes, and the likeliest confound is obvious to any teacher: students who are struggling reach for help. The same round of results found one in five 15-year-olds is now a low performer in science, mathematics and reading, up from 16 percent in 2022, with the lowest OECD-average performance recorded so far. A board handed the headline will reach for prohibition. A board handed the actual sentence will ask a better question, which is what students are being taught to do with the thing rather than whether they touch it.
None of this means waiting for someone else to decide. In evidence to the Commons Education Select Committee reported by Tes on 9 September 2026, the National Institute of Teaching's Pip Sanderson described leaders as "desperate for really clear guidance and advice around what to do," and the NASUWT's Darren Northcott called the Department for Education's approach "laissez-faire," asking for "much clearer rules of the road." Whatever arrives, it will arrive after your next board meeting. Sanderson's summary of the mood is the sentence to take into that meeting: "This is all great and really interesting, but what do I do?"
What to do before your next governing body meeting
Do four things, and none of them requires new work.
Open last term's AI paper and run the three questions of the Governors' Test over it. Be honest about the first one. Then write this term's paper as four lines and put everything else in an appendix. Then find the sentence in your own school that answers "who tells the board if this changes before the next meeting," and if there is no sentence, that is your one decision for this term. Finally, name the watch list: the two or three things you would want to know had gone wrong, so that line three has something to report even in a quiet term.
If your board has never been given a position it could hold you to, this term is a good time to give it one. Start with seven dimensions worth assessing honestly before you write the paper, be clear about who answers when an AI tool gets something wrong, and if the paper keeps arriving late or thin, the problem may be that nobody owns it, which is a question about who sits on your AI steering group. If the decision in front of your board is a restriction, make it a decision rather than a mood. And if none of it adds up yet, the underlying job is what responsible AI adoption looks like in practice.
If your leadership team is working out what its board actually needs to see, this is the kind of work I support through AI governance and strategy sessions with leadership teams, boards and trustees. I also write about these questions daily, and you can get that in your inbox.
Sources and further reading
- OECD, PISA 2025 Results (Volume I): Future-Ready Students, executive summary, launched 8 September 2026. Read it
- OECD, "PISA 2025: Students' reading and mathematics performance declined sharply across the OECD," press release, September 2026. Read it
- IBM and Morning Consult, AI Readiness in US Schools, 2 September 2026 (fieldwork July 2026; 1,019 K-12 education professionals and 1,029 parents; margin of error 3 percentage points). Read it
- Department for Education, Keeping Children Safe in Education 2026, statutory guidance for schools and colleges, September 2026. Read it
- Office of the Mayor of New York City, "Mayor Mamdani and Chancellor Samuels Put Students First with Nation's Broadest Generative AI Moratorium in Schools," 2 September 2026. Read it
- Jabed Ahmed, "Schools 'desperate' for clearer AI guidance," Tes, 9 September 2026. Read it
Dan Fitzpatrick is the founder of The AI Educator, a Forbes contributor and a bestselling author on AI in education. More about Dan.
