G7 Day One: frontier AI CEOs are in Évian — and Fable 5 is still offline

The 52nd G7 Summit opened this morning in Évian-les-Bains, France — and for the first time, all three major frontier AI lab CEOs are on-site: Sam Altman (OpenAI), Dario Amodei (Anthropic), and Demis Hassabis (Google DeepMind). Macron personally invited Altman, marking his first G7 appearance. The three co-signed a letter to Congress on AI biosecurity risks earlier this month — an unusual alignment that shapes the tone they bring to the table.

An irony the summit can't ignore. Amodei arrived while his company's two most capable models are offline under a US government export control directive, simultaneously invited to co-shape global AI governance with the heads of state who wield that regulatory power. The Fable 5/Mythos 5 shutdown — covered in our June 13 post and yesterday's Amazon backstory piece — is now a live concrete example of what unilateral government model recall looks like without a structured pre-release framework.

The AI agenda. A dedicated AI working lunch is scheduled for Wednesday, June 17, focused on "ensuring safe, rapid, and effective AI deployment." Per Dataconomy, OpenAI expects companies to leave having agreed to voluntary AI safety commitments, with youth safety and frontier AI risks (cyber and biological) as the primary themes.

What to expect — and not expect — from the communiqué. The Trump administration has signaled resistance to binding multilateral AI constraints that could limit US competitiveness, per TechPolicy.Press. Expect "responsible use" framing and voluntary commitments rather than mandatory pre-release certification in the final text.

Why builders should track this. G7 communiqués don't make law — but they set the framing that does. Any reference to coordinated "frontier model evaluation" or "pre-release reporting requirements" in the June 17 closing text is worth reading carefully. If the Fable 5 incident ends up cited as a case study for why an established international pre-clearance process is needed, the next model you deploy into production could be subject to rules shaped by what happened this week.

Sources: EU Council: G7 summit June 15-17, Euronews: What to expect from G7, Dataconomy: AI leaders at G7, The Next Web: AI rivals head to G7, TechPolicy.Press: rift over AI sovereignty