Kia ora from New Zealand!
Sorry – long flight got super extended (9.5 hour layover, then a redirect to Auckland due to bad weather) so it ate away the time I'd meant to spend putting the newsletter together. I'm heading into ATAIN/TICT26 in Wellington, so this one's a special edition rather than the usual five.
Two things instead. Earlier this month Times Higher Education ran a piece of mine responding to Dario Amodei's essay on AI governance. The opening half is below in my own words rather than the sub-edited version, and the link to the full piece is in the middle. And this week's episode turned out to be about exactly the same problem one floor down, which I hadn't planned and which is the reason both are in here together.
Short version of the pair: nobody in the building can see what anyone else is actually doing.
Right Levers, Wrong Building: What Dario Amodei Got Right and Where His Argument Stops | Sensing Failure 🕐
Four days. Tuesday, Anthropic releases Fable, its most powerful public AI model. Wednesday, its CEO publishes a landmark essay calling for government oversight of frontier AI. By Friday, the US government has suspended access to that model for foreign nationals, citing national security concerns, forcing Anthropic to take it offline before the week is out.
Every organisation that had signed up to use these tools closed out the week with less than it had on Tuesday morning.
Amodei's central argument – that AI moves in weeks - while policy moves in years, and that the mismatch is the defining governance challenge of our moment – is exactly right. But his solutions stop at the institution's front door. He's writing about nation-states and regulatory bodies: FAA-style model testing, export controls, democratic coalitions. Right levers, wrong institution for most people reading this.
He has nothing to say about what happens inside universities, where EDUCAUSE's 2026 survey of nearly 2,000 US higher education staff found 94% using AI tools for work while 46% were unaware of any institutional policy governing that use – including 38% of executive leaders. In many cases, the policies simply don't exist.
The argument I make in the piece is that this isn't a speed problem. Three clocks – model, institution, practitioner – run inside every university at permanently mismatched speeds, with no transmission between them. They cannot be synchronised. Speed up governance and you sacrifice the deliberation that makes its decisions legitimate. Suppress the practitioner clock and you lose the most honest signal the institution has. And the model clock will not slow for a working group. Call it the three-clock problem.
That's the diagnosis. The rest of the piece is the harder half – what any individual can do that a properly constituted committee cannot, why a sensing function doesn't undermine deliberation, and the small, specific ask I'd put to anyone with a governance seat.
📰 Read the rest in Times Higher Education:
The Kids Are All Right: Students Are Writing Their Own AI Rules | Adjunct Intelligence 🎙️
Dale Leszczynski and I spent this week's episode on what students actually said, rather than talking for them. The line that stuck was from a 16-year-old who wrote a breakup letter to ChatGPT for a New York Times student writing contest:
"Your voice started to replace my own."
She named the harm better than most of the literature has, and then cut herself off. Nobody made her.
There's data underneath it. A Georgetown preprint looked at 370,000 college application essays, before and after ChatGPT. Human judges rate the later ones more creative – with up to eight times fewer original ideas in them. Better surface, thinner thinking.
And it doesn't land evenly. The voices furthest from the average get smoothed toward the middle hardest: neurodivergent students, and students writing across languages. Which, from where I sit in Vietnam, means the same student gets pushed toward sounding like a machine by one system and penalised for sounding like a machine by another. Same semester, taxed twice.
Nobody designed that. It's a second-order effect, and second-order effects are what you get when nobody is watching the whole loop – the argument in the piece above, arriving from the other end of the building.
The students are ahead of us on it. One in a Jisc survey had written her own policy in a sentence: anything on her Word doc has to be her own sentences. Nobody told her to. No credit for following it, no penalty for breaking it.
They're doing their half. Integrity was never a student attribute – it's a design outcome, and the design is ours.
🎧 Students Are Writing Their Own AI Rules – Adjunct Intelligence, out now:
🎓 Higher Education Horizons 2026 – The Human Edge – RMIT Saigon South, 24–25 September, free, Tim Fawns keynoting. Registration is open and it fills. #shamelessplug
Normal service next Monday, from a desk, with a full pool of articles.