For the First Time, the Three Rival AI Lab CEOs Stood Before G7 Leaders Together

Évian-les-Bains is best known for its spring water. This week, it became the setting for something more consequential: the first time the chief executives of the world’s three leading AI laboratories sat across from heads of state at the same table.

Sam Altman of OpenAI, Dario Amodei of Anthropic, and Demis Hassabis of Google DeepMind traveled to France’s Lake Geneva resort town for the 52nd G7 Leaders’ Summit, which ran June 15–17 under the French presidency of Emmanuel Macron. They were joined by a broader contingent of AI executives — including Arthur Mensch of Paris-based Mistral AI, Aidan Gomez of Cohere, Alex Wang of Meta, Marc Benioff of Salesforce, and Ren Ito of Sakana AI — but it was the gathering of AI’s acknowledged “big three” that drew the most geopolitical attention.

A Governance Moment That’s Been Years Coming

The optics alone were striking. Just weeks before the summit, OpenAI and Anthropic each made confidential S-1 filings with the U.S. Securities and Exchange Commission, signaling imminent public listings that will expose both companies to far greater scrutiny of their operations, governance, and safety practices. For Altman and Amodei, the G7 appearance arrived at a moment when managing government relations has never been more commercially sensitive.

The summit’s AI agenda was built on the architecture of the Hiroshima AI Process, launched at the 2023 G7 meeting in Japan. That initiative produced a set of international principles and voluntary codes of conduct. Canada’s 2025 presidency deepened commitments around AI adoption in public services and online youth safety. France’s priorities as current chair include AI security, encouraging AI adoption among small and medium-sized enterprises, sustainable technological development, and protection of children online.

What the process has not produced — critics note bluntly — is anything enforceable. The working assumption among G7 governments remains that binding regulation is too blunt an instrument for a technology evolving at AI’s pace, and that the laboratories themselves must remain cooperative partners rather than adversarial compliance targets.

What the CEOs Said — and What They Didn’t

Spokespersons for both OpenAI and Anthropic confirmed in advance that the summit would address “the opportunities and threats posed by advanced AI,” but neither company provided a detailed agenda. Google DeepMind declined to elaborate beyond confirming Hassabis’s attendance.

The reticence was itself revealing. Frontier AI risks — particularly in the cyber and biological domains — are cited by summit organizers as a key area of focus. Yet the companies most capable of producing such risks are understandably cautious about the specificity of what they commit to in a multilateral setting where any statement can become a negotiating anchor.

One theme that cut across multiple delegations was youth safety. Altman has made children’s online protection a stated personal priority in his public remarks throughout 2026, and Canadian Prime Minister Mark Carney separately pressed for AI safety standards that would protect children’s data from exploitation. The convergence created at least one area of near-consensus — even if the mechanisms for enforcement remain contested.

A Structural Tension the Summit Cannot Resolve

The deeper problem on display at Évian is one that no communiqué can solve: the pace at which frontier AI capabilities are advancing has consistently outrun the pace at which governance frameworks can be designed, negotiated, and ratified.

Claude Fable 5, released earlier this year and already the subject of ongoing talks between Anthropic and the White House, was described by some national security officials as presenting unprecedented dual-use risks. OpenAI’s next generation of models — expected to push further into autonomous reasoning and real-world action-taking — are approaching release. The models being discussed this week at the policy level already represent yesterday’s frontier.

Against that backdrop, voluntary pledges look increasingly inadequate to critics who argue that frontier AI has crossed into territory requiring the kind of binding international coordination historically applied to nuclear or biological weapons programs. European voices at the summit — including those from EU member states already operating under the AI Act’s enforcement regime — pushed for more concrete commitments that the United States is not yet prepared to make.

The American counterargument, advanced consistently across the Trump administration’s various AI policy pronouncements, is that over-regulation risks ceding the frontier to less safety-conscious actors — primarily China. The Hiroshima Process was designed in part to develop governance frameworks among democratic allies while excluding Beijing, a goal that remains structurally intact but increasingly strained by the rapid capability gains of Chinese frontier labs like DeepSeek and Moonshot AI.

The Peculiar Position of the Lab CEOs

There is an inherent tension in Altman, Amodei, and Hassabis appearing before governments as de facto AI governance partners. Each of them runs a company that has publicly acknowledged — in safety documentation, congressional testimony, and regulatory filings — that the technology they are building could pose catastrophic risks if misaligned or misused.

Yet each of them is also racing to build more powerful systems on timelines that preclude the kind of careful, iterative governance review that civil society groups are demanding. The G7 setting allows them to present as responsible actors willing to engage with governments without making enforceable commitments that would constrain their development velocity.

The simultaneous IPO preparations are not incidental to this dynamic. Going public will require OpenAI and Anthropic to disclose far more about their financials, risk factors, and governance structures than they ever have as private companies. Cultivating relationships with G7 heads of state is one way to build the institutional credibility those disclosures will require.

What Comes Next

The summit formally closed on June 17. But the follow-on process is already scheduled: the data protection authorities of all seven G7 nations will convene in Paris on June 23 for a four-day roundtable on AI enforcement, cross-border data flows, and what regulators describe as a long-exploited jurisdictional loophole that has allowed AI companies to process data in ways that circumvent national privacy laws.

That meeting will be chaired by France’s Commission Nationale de l’Informatique et des Libertés. Whether it produces anything more concrete than Évian remains to be seen. But for the first time, the people building frontier AI and the governments responsible for governing it are at least in the same room.

The distance between those two groups, however, remains considerable. And AI capability development isn’t pausing to let the governance gap close.