Altman, Amodei, Hassabis at G7: Europeans Push Back on US AI Dominance

The most unusual meeting at the G7 Évian summit was not between heads of state. It was a working lunch on June 17 where Trump, Macron, Scholz, Meloni, Sunak, Trudeau, and the European Commission President sat across from approximately 15 chief executives of the world's most powerful AI companies. The formal theme was "Ensuring a safe, rapid and effective deployment of artificial intelligence." The actual debate was whether the rest of the world can accept that America's AI companies are increasingly operating under conditions defined by the US executive branch.

Who Was in the Room

The confirmed attendees from the AI industry include Sam Altman (OpenAI), Dario Amodei (Anthropic), Demis Hassabis (Google DeepMind), Arthur Mensch (Mistral AI), Aidan Gomez (Cohere), Robin Rombach (Black Forest Labs), Pratyush Kumar (Sarvam AI), Victor Riparbelli (Synthesia), Marc Benioff (Salesforce), Ren Ito (Sakana AI), and Alex Wang (Meta).

The group is geographically diverse by design. Mensch is French, representing the leading European frontier AI company. Kumar leads Sarvam AI, India's most prominent AI foundation model company. Rombach leads Black Forest Labs, a German image generation company. The inclusion of non-G7-country CEOs in a G7 working lunch is itself a statement that AI governance is already a global problem that does not align with the G7's membership.

Notably absent: any Chinese AI company. The working lunch composition reflects the current reality of AI geopolitics, where the conversation about safety and governance is happening primarily among US, European, and allied-nation companies.

The Anthropic Incident That Changed the European Conversation

The single most politically consequential AI event before the summit was Anthropic's removal of its Fable 5 and Mythos 5 models from global access. The Trump administration issued an order citing unspecified national security concerns, and Anthropic complied by suspending access to both models for all non-American users, regardless of whether they were inside or outside the United States.

The removal affected European developers, research institutions, and enterprises using Anthropic's API. It happened without advance notice and without public explanation of the specific security concern. European AI policy officials described it as precisely the scenario they had been warning about for two years: a US AI company, under US executive branch direction, unilaterally withdrawing AI infrastructure from global users with no recourse.

Fable 5 is Anthropic's most capable model as of mid-2026. Its sudden unavailability to non-American users disrupted production workflows across thousands of companies and research groups in Europe, Japan, and beyond. The incident accelerated existing conversations about AI sovereignty from theoretical to urgent.

What Europeans Want: Sovereignty Without Isolation

European leaders at Évian are not seeking to build AI systems that exclude American models. What they want is assurance that access to foundational AI infrastructure cannot be revoked unilaterally by the US executive branch.

The European Commission unveiled a tech sovereignty package in June 2026 that includes plans to boost homegrown AI, reduce dependency on US hyperscaler compute, and establish contractual protections for European users of American AI services. The package includes proposals for data residency requirements, interoperability mandates for frontier AI APIs, and minimum notice periods before service withdrawals.

Macron's position at the summit is to frame this as complementary rather than adversarial to US interests. France hosts Mistral AI, which has positioned itself as a European alternative to American frontier models. Mistral is also increasingly working on enterprise sovereign deployments where model weights are hosted entirely within European infrastructure and access is not dependent on any US company's continued commercial cooperation.

Cohere's Aidan Gomez stated the company's summit position directly: the goal is "to expand our sovereign AI ecosystem partnerships beyond Canada and Germany to include all G7 nations." Cohere, which acquired German AI startup Aleph Alpha in early 2026, operates sovereign deployment infrastructure where governments and enterprises can run Cohere's models without the data or compute transiting US jurisdiction.

Frontier AI Risks: What the Working Lunch Actually Covered

The summit working lunch agenda covered three areas: frontier AI risks in the cyber and biological domains, AI infrastructure and sovereignty, and regulatory coordination across G7 jurisdictions.

The frontier AI risk discussion centered on two scenarios. First, adversarial use of AI for cyberattacks, specifically the use of frontier models to automate vulnerability discovery at scale and to generate novel malware beyond current defensive detection capabilities. Second, the potential for AI to accelerate biological threat development, including the use of large language models to assist in designing pathogen variants or synthesis routes for dangerous compounds.

Both areas are treated by the US intelligence community as active threat vectors where defensive AI capabilities need to match or exceed offensive development. The working lunch discussion included briefings from national security advisors alongside the AI CEOs, which is itself new in G7 history.

The outcome on frontier AI risks was a shared statement that governments and leading AI developers should establish incident reporting protocols for cases where AI systems are misused in cyber or biological contexts. No specific enforcement mechanism was agreed, but the statement was more concrete than the 2023 Hiroshima AI process's voluntary principles.

The EU AI Act Simplification Running Simultaneously

While the G7 working lunch was happening, a separate regulatory development was in motion. The European Council and Parliament agreed on May 7, 2026 to simplify and streamline the EU AI Act. The simplification removes some compliance layers for general-purpose AI models, reduces bureaucratic burden for smaller EU AI companies, and adjusts the timeline for high-risk AI system provisions to August 2, 2026.

The simplification reflects two years of industry feedback that the original AI Act compliance requirements were disproportionately burdensome for smaller European AI companies relative to large US hyperscalers that could absorb compliance costs more easily. The revised framework focuses compliance burdens on deployers and applications rather than foundation model providers in most cases.

For developers, the EU AI Act simplification means the compliance stack for deploying AI applications in Europe is slightly less complex than originally designed. The August 2, 2026 effective date for high-risk AI system provisions is the most important near-term deadline for companies building AI tools in regulated sectors (healthcare, finance, education, employment screening, critical infrastructure).

The G7 AI Governance Statement: What Was Agreed

The G7 digital and technology ministers agreed in late May on priorities covering secure AI, AI openness, digital sector resilience, and online child safety. The heads of state summit in Évian added a geopolitical dimension: a statement that AI infrastructure should be built on principles of interoperability and that no G7 nation's developers should face unilateral loss of access to AI infrastructure without due process.

The statement does not name the Anthropic Fable 5 incident explicitly. But the language about "due process" and "advance notice" in AI service withdrawal is directly responsive to it.

The AI sovereignty section of the G7 communiqué also endorses the concept of "AI interoperability standards" to be developed in coordination with the OECD and ISO. These standards would define minimum technical requirements for frontier AI APIs such that a model accessible in one G7 jurisdiction meets baseline technical compatibility with applications built in another.

For developers, interoperability standards reduce vendor lock-in risk. If your application is built on an OpenAI-compatible API schema, interoperability standards would require that equivalent Mistral or Cohere endpoints work without significant code changes. This is aspirational in 2026 but the G7 endorsement means it enters the formal standards development pipeline.

Our Analysis: The Sovereignty Gap Is Widening, Not Narrowing

The Anthropic Fable 5 incident is a preview of a structural problem. As US AI companies build more capable frontier models and as those models become embedded in critical infrastructure worldwide, the US executive branch's ability to direct their global access creates asymmetric risk for every non-American user.

The G7 working lunch is a political acknowledgment of this problem. It does not solve it. Mistral remains a distant second to OpenAI and Anthropic on benchmark performance. Cohere's sovereign deployments work for enterprise use cases but do not reach consumer or developer-accessible frontier performance. Sarvam AI and Sakana AI are regional specialists.

The structural response requires either: US legal protections that limit executive branch power over commercial AI access for allied-nation users; or European and allied-nation frontier model development that reaches genuine capability parity. Neither is happening fast enough to close the gap before the next access revocation incident.

What is happening is that the cost of dependency is becoming visible. Anthropic's Fable 5 removal cost European enterprises real money and disrupted real production systems. That cost is now being priced into procurement decisions. Enterprises in Europe and Japan are increasingly adding sovereign deployment requirements to AI vendor contracts as a condition of purchase. This is the market pressure that the G7 political framework is formalizing.

See the G7 Évian summit overview, the Anthropic $96.5B valuation analysis, and the CXMT $8B memory chip investment for related context.

Key Takeaways

• 15 AI CEOs attended the G7 Évian working lunch including Altman (OpenAI), Amodei (Anthropic), Hassabis (Google DeepMind), Mensch (Mistral), Gomez (Cohere), Benioff (Salesforce), and Wang (Meta)
• Anthropic blocked Fable 5 and Mythos 5 from all non-Americans following a Trump administration national security order, the immediate trigger for Europe's sovereignty demands at the summit
• Cohere acquired Aleph Alpha in early 2026 and is positioning sovereign G7 AI deployments where governments can run models without US compute or data transit dependency
• EU AI Act simplification agreed May 7, 2026: high-risk AI system provisions take effect August 2, 2026; compliance burden reduced for foundation model providers, shifted to deployers and applications
• G7 AI governance statement endorses interoperability standards: OECD and ISO process begins; goal is that frontier AI APIs are compatible across G7 jurisdictions to reduce vendor lock-in risk
• Frontier AI risk protocols agreed: incident reporting for AI misuse in cyber and biological domains, more concrete than the 2023 Hiroshima process voluntary principles

Sources

• CNBC — AI in spotlight at G7 as Trump, world leaders joined by tech chiefs
• Washington Post — AI executives gather at G7 as Europeans seek checks on American dominance
• Fortune — The G7 has some special lunchtime guests this year: Sam Altman, Demis Hassabis and Dario Amodei
• The Hill — G7 summit focuses on contentious future of AI and US dominance of the industry
• European Commission — G7 Digital and Technology Ministerial Declaration May 29, 2026