At a closed-door lunch during the G7 summit in Évian-les-Bains, France, Google DeepMind CEO Demis Hassabis, alongside Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman, advocated for a US-led international coalition to establish rules and standards for artificial intelligence. Hassabis proposed a technical standards body that would adapt its guidelines quarterly as new risks from AI systems emerge. He emphasized the need for an international framework to certify AI models, akin to a "kite mark of quality," ensuring they meet safety standards and prohibiting characteristics like deception.
Hassabis's proposal suggested national AI safety institutes, similar to the one in the UK, would feed into a global body responsible for defining benchmarks for undesirable AI properties. He specifically flagged deception and AI systems communicating in non-human-readable tokens as vulnerabilities that should be banned. This initiative aims to address two core risks: misuse of dual-purpose AI technologies by malicious actors and ensuring increasingly autonomous AI systems remain within developer-set guardrails, especially as these are expected to emerge within a year or two.
While Hassabis called for a US-led approach to increase trust and provide a certification process, other AI leaders present had differing views. Dario Amodei of Anthropic also called for a US-led coalition of democratic countries to control access to technology and isolate "common adversaries" such as China. In contrast, Sam Altman of OpenAI underscored the importance of an "international forum for discussion that establishes globally accepted standards for testing" but cautioned against concentrating too much power in the hands of a few tech companies, including his own. He stressed that governments should not cede their responsibilities to AI labs.
Despite the discussions among top AI executives and G7 heads of state, including President Donald Trump, the gathering yielded no binding commitments or regulatory announcements. The meeting was largely described as a conversation rather than a negotiation, with several G7 members, notably France and the EU, pursuing their own AI regulatory paths that do not necessarily presuppose American leadership. The call for a US-led coalition came just days after the Trump administration imposed export controls on Anthropic's Fable 5 and Mythos 5 models due to national security concerns.
