Picture the world's most powerful leaders sitting down to a working lunch, and three AI CEOs pulling up chairs to explain, between courses, that their own creations might be a teensy bit dangerous. That actually happened this week, and the salad came with a generous side of geopolitics.

Three CEOs, One Very Loaded Lunch

On June 17, 2026, OpenAI's Sam Altman, Anthropic's Dario Amodei, and Google DeepMind's Demis Hassabis joined Trump and G7 leaders for a working AI lunch in Évian-les-Bains, France. Amodei and Hassabis used the moment to call for a U.S.-led coalition of democratic countries to set common standards for developing, evaluating, and governing advanced AI.

Amodei named specifics: structured access to frontier models, chip trade agreements that pointedly exclude China, and cooperation on the risks of AI in cyber, bioterrorism, and intelligence. Altman took a softer tack, pitching an international forum to establish globally accepted testing standards and provide impartial analysis of AI's capabilities and risks.

Safety Pitch, Sovereignty Subtext

It's a remarkable bit of theater: the people building the most powerful AI systems on Earth flying to a summit to warn governments those very systems are too dangerous to leave ungoverned. Cynics call it regulatory moat-building; optimists call it unusual candor. The truth is probably sitting right there at the lunch table, eating both.

Strip away the diplomatic seasoning and this is as much about market access as safety. A "democratic coalition" with shared standards and China-excluding chip deals is also, conveniently, a coalition that runs on American models and American silicon. The timing makes the subtext loud: the summit came just days after Washington blocked all foreign access to two frontier models — reportedly the first time the U.S. applied export controls to an AI model directly.

Nothing says "global standards for the good of humanity" quite like making sure the standards run on your stuff.

Source: CNBC