Anthropic’s Dario Amodei and Google DeepMind’s Demis Hassabis used a private lunch on the G7’s last day to press the United States into leading a new international AI coalition. The gathering, held in the French spa town of Évian‑les‑Bains, brought together about a dozen senior executives from the world’s leading AI labs alongside the heads of the Group of Seven nations.

Amodei outlined a roadmap that would give participating countries structured access to frontier models while barring China from critical chip and component trade. He also warned that unchecked AI could be weaponized in cyber attacks, bioterrorism and intelligence‑gathering operations. Hassabis echoed the call, emphasizing the need for shared standards that balance innovation with security.

OpenAI’s Sam Altman, present at the lunch, broadened the discussion. He advocated for an “international forum for discussion” that would establish globally accepted testing standards, provide impartial risk analysis, and serve as a venue for cross‑border cooperation. OpenAI’s global‑affairs chief Chris Lehane noted that non‑U.S. leaders in the room recognized Washington’s capacity to steer AI governance.

Canadian officials, representing Prime Minister Justin Trudeau, responded positively, indicating that Canada would welcome a U.S.‑led coalition. The Canadian side did not elaborate further, and neither Anthropic nor DeepMind issued immediate comments after the meeting.

Despite the high‑level attendance, the lunch yielded no concrete commitments. Participants described the session as a conversation rather than a negotiation, and the G7’s track record on AI—ranging from the Hiroshima AI Process to Canada’s 2025 presidency pledges—has so far produced only principles and voluntary codes, not enforceable regulation.

The timing of the coalition pitch is significant. Just five days earlier, the Trump administration imposed export controls on Anthropic’s flagship Fable 5 and Mythos 5 models, citing national‑security concerns after a reported jailbreak. Anthropic disabled the models worldwide to comply, and talks with the Commerce Department continued into the following week. The move underscored the tension between AI firms positioning themselves as strategic partners to Washington and the government’s willingness to act unilaterally against perceived risks.

Other tech leaders at the lunch included Arthur Mensch of France’s Mistral, Aidan Gomez of Cohere, Robin Rombach of Black Forest Labs, Victor Riparbelli of Synthesia, Salesforce’s Marc Benioff, and Meta’s Alex Wang. Executives from Italy’s Domyn, India’s Sarvam AI and Japan’s Sakana AI also participated, reflecting France’s effort to frame the AI dialogue as a truly global conversation rather than a U.S.–centric one.

The cybersecurity angle loomed large. OpenAI recently released a limited preview of GPT‑5.5 Cyber to vetted security teams, while Anthropic’s Mythos model had been restricted to defenders before the export controls forced a full shutdown. Both companies argue that frontier AI tools are more valuable to defenders than attackers, a stance complicated by the U.S. government’s recent actions.

Whether a U.S.-led coalition will materialize remains uncertain. The Trump administration’s willingness to enforce export bans could undermine collaborative frameworks, while European allies such as France continue to pursue independent regulatory paths. The lunch, however, made clear that AI CEOs now see shaping international governance as essential to their businesses’ future, not merely a regulatory burden.

Este artigo foi escrito com a assistência de IA.
News Factory APP - notícias agênticas para impulsionar seu SEO e AEO.