G7 Leaders Meet AI CEOs in Évian, Pledge Safer Chatbots for Children

Leaders of the Group of Seven closed their summit in Évian-les-Bains, France, on June 17 after a working lunch with the heads of the biggest AI developers. They pledged to work with leading companies to speed up the safe deployment of AI and to make chatbots safer for children, but adopted no standalone AI declaration. For enterprise leaders, the summit offered a clear read on where governments are heading: toward child-safety rules for conversational AI, common standards for evaluating models, and growing worry about depending on a handful of providers.

Who attended the G7 AI discussion in Évian?

The working lunch on June 17 was titled "Ensuring a safe, fast and efficient deployment of artificial intelligence," according to the list of expected business participants published by the G7 Research Group at the University of Toronto. The 12 expected executives included Sam Altman of OpenAI, Dario Amodei of Anthropic, Demis Hassabis of Google, Alexandr Wang of Meta, Marc Benioff of Salesforce, Arthur Mensch of Mistral AI and Aidan Gomez of Cohere. The list also featured leaders from Sakana AI in Japan, Black Forest Labs in Germany, Synthesia in the UK, Domyn in Italy and Sarvam AI in India, alongside partner countries Brazil, India, Kenya, South Korea and Egypt. The group listed them as expected, so attendance by each was not confirmed in that document.

What did G7 leaders agree on AI?

The French presidency's summary of outcomes said members are working together with leading companies to accelerate the safe and beneficial deployment of AI. It also said leaders agreed to protect young people from AI risks, including by adapting the language chatbots use with children.

The detail is in the Leaders' Call on a Safer Digital Space for Minors, also dated June 17. It asks providers of conversational AI to build safety settings by default for children and youth, including parental controls and age assurance. It calls on platforms to use privacy-preserving age assurance, reduce engagement-maximizing features, and support more reliable and interoperable tools for identifying synthetic content. Governments said they would support common standards for evaluating AI models and algorithmic systems, and ministers are to assess progress by the end of the year.

The G7 Research Group's index of summit documents lists statements on growth, critical minerals, partnerships and other topics, but no separate leaders' declaration on artificial intelligence.

How did the Anthropic model suspension shape the summit?

The meeting followed a US export control directive on June 12 that led Anthropic to disable its newest models, Fable 5 and Mythos 5, for all customers. On June 14, Canadian Prime Minister Mark Carney said the episode showed the risk of relying on a few powerful AI models and compared it to concentration risk in the 2008 financial crisis, The Next Web reported. He said that "nobody's done anything wrong in this situation" and that Washington had identified some risks with the models. He also lowered expectations for the summit itself.

Carney's government launched a national AI strategy on June 4 that includes sovereign computing infrastructure and a goal of raising business AI adoption in Canada from 12% to 60% by 2034, according to the same report. Other governments have moved in a similar direction: The Next Web noted an EU tech sovereignty package earlier in June and a British effort backed by BT, HSBC and BAE to build a sovereign frontier model.

The trade-off is real on both sides. National capacity can reduce strategic dependence, while the most capable models and the largest cloud platforms are still concentrated among a few US providers. The summit outcomes did not attempt to resolve that tension.

Practitioners at AI Connect events have framed the same question for enterprises. NVIDIA's Shilpa Kolhatkar, Global Head of AI Nations, said during a AI Connect webinar that with adoption everywhere, "the critical question is now who is producing it?" Our coverage of sovereign AI and air-gapped deployments explains what that means for data, operations and technology control.

What should enterprise leaders take from the G7 summit?

The summit did not create new binding rules, but it points to obligations that customer-facing AI teams can plan for now:

• Child safety by default. Companies that run chatbots consumers can reach should review age assurance, default settings and parental controls before regulators in G7 countries translate these commitments into requirements.
• Content provenance. Support for interoperable synthetic-content detection suggests labeling and provenance tools will become an expected control.
• Evaluation standards. Common model evaluation standards would give buyers a shared yardstick. Until then, enterprises should keep their own tests, as AI Connect speakers argued in our piece on custom AI benchmarks.
• Provider concentration. The policy conversation about dependence on a few models is now at head-of-government level. Boards will ask about it, and governance teams should have an answer, a theme we explored in AI data governance at speed.

The next checkpoint is the ministerial progress review on minors' online safety promised by the end of the year.

Frequently asked questions

What did the G7 agree on AI in Évian?

Leaders said they would work with leading companies to speed safe and beneficial AI deployment and agreed to make chatbots safer for children, including default safety settings and age assurance.

Which AI executives were invited to the G7 summit?

The expected list included Sam Altman, Dario Amodei, Demis Hassabis, Alexandr Wang, Marc Benioff, Arthur Mensch and Aidan Gomez, among 12 business leaders.

Does the G7 outcome create binding AI rules?

No. The commitments are political, with ministers asked to assess progress on minors' online safety by the end of the year.

Filed under: AI Governance & Trust, AI Strategy & ROI