The G7 Set a Lunch Table for AI's CEOs. They Pitched a Wall
At Evian, Amodei and Hassabis asked heads of state for a US-led AI coalition built on chip controls that exclude China — while Europe, watching one American model sit export-banned, pushed for sovereignty instead.
The 52nd G7 summit in Evian did something summits rarely do: it pulled the people who build frontier AI into the same room as the people who are supposed to govern it. On Wednesday, June 17, 2026, a working lunch themed "Ensuring a safe, rapid and effective deployment of artificial intelligence" sat Sam Altman of OpenAI, Dario Amodei of Anthropic, Demis Hassabis of Google DeepMind, Arthur Mensch of Mistral, and roughly a dozen other tech leaders down with heads of state, including President Trump and host President Macron. The optics were unusual. The asks were more revealing than the optics.
A coalition, not a referee
Read coverage of these moments and you expect the CEOs to play defense — to fend off regulation, soften timelines, promise self-governance. That isn't what happened at Evian. Amodei and Hassabis went on offense, calling for a U.S.-led AI coalition and pitching the heads of state on an international forum — potentially helmed by the United States — to set global standards for advanced models.
The framing matters. The stated goal of those standards was not primarily safety in the abstract; it was, in the executives' own telling, ensuring "ongoing and continued access to the frontier models." In other words: a governance structure whose first job is to keep the frontier flowing to the right countries. Amodei was specific about what the coalition should coordinate — structured access to frontier models, and the trade of chips and critical components, explicitly excluding China. This is industrial policy wearing a safety lanyard. The proposal isn't a neutral referee for a global technology; it's a bloc, with a membership line drawn through Beijing.
Why this week, of all weeks
The timing turned the pitch from ambitious to pointed. The lunch happened on day six of an export-control directive that had pulled Anthropic's own frontier model, Fable 5, offline — a model suspended in a dispute that began when a Korea-based carrier was flagged as a China-linked security risk. So the man arguing for a chip-and-model wall against China was doing it while watching his flagship sit dark on the wrong side of an American control regime that had just bitten him.
That is not the contradiction it looks like. It's the argument. Amodei's case is essentially that the controls are coming either way, messily and unilaterally, so the labs would rather they be coordinated, predictable, and built around a coalition that guarantees members keep their access — instead of the current world, where a single directive can black out a frontier model overnight and leave even a U.S. company stranded. The Fable 5 blackout wasn't a counterexample to the pitch. It was the exhibit.
Europe wants a lock, not a key
The other half of the table heard something different. To several European participants, "U.S.-led coalition" and "continued access to frontier models" translated to a permanent dependency on American companies and American export decisions — the very arrangement they came to Evian to resist. Arthur Mensch, representing Mistral and, by extension, the case for European AI sovereignty, embodied the pushback: Europe seeking checks on American dominance, not a seat in an American-run club.
It's a genuine fork, not a misunderstanding. The labs are offering Europe a key — guaranteed access to the best models, provided it joins the bloc and accepts the chip controls that come with it. Europe increasingly wants a lock — its own models, its own compute, its own ability to say no — even if that means running a step behind the frontier for a while. A model that can be switched off by a foreign government's directive, as Fable 5 just was, is precisely the future European officials are trying to engineer their way out of. Watching an American model get blacked out by Washington didn't reassure them about access guarantees; it confirmed their fear.
What actually got agreed
Concrete outcomes were thinner than the rhetoric, which is normal for a summit. The leaders coalesced around a voluntary commitments framework touching frontier AI risk and the protection of children online — the Élysée flagged youth safety as a central thread of the discussion. "Voluntary" is doing heavy lifting there; it's the register of a group that agrees something should be done and disagrees about who should be in charge of doing it.
But the durable takeaway from Evian isn't the framework. It's the posture. The frontier labs have stopped asking governments merely to permit them and started asking governments to organize the world around them — to build the trade rules, the access tiers, and the coalition membership that would harden their lead into geopolitical infrastructure. They pitched it at the highest table available, on the same week one of their own models proved how fragile that access still is.
The G7 set a lunch table for AI's CEOs expecting to question them. The CEOs showed up with a blueprint for a wall, a key for their friends, and a live demonstration — courtesy of a banned model sitting offline a few time zones away — of exactly what they were trying to prevent. Europe, holding the key it was being offered, decided it would rather forge its own lock.