Five days into the US-ordered shutdown of Anthropic’s Fable 5 and Mythos 5 models, the story has moved from Washington to the French Alps. On the sidelines of the G7 Leaders Summit in Évian-les-Bains, allied governments are actively pushing for an exemption framework that would restore access to US frontier AI models for vetted countries and companies.

What Happened at the Summit

On the evening of Monday, June 16, G7 leaders held informal discussions about a proposed “trusted partners” scheme that would allow select foreign governments and enterprises to regain access to Anthropic’s most capable models, which were pulled from global availability on June 12 under a US Commerce Department export control directive.

Three diplomatic sources confirmed the discussions took place primarily on the sidelines of the summit opening dinner, with US Commerce Secretary Howard Lutnick the key American counterpart in those talks. European leaders reportedly pushed Lutnick directly, calling on Washington to reconsider the blanket ban and replace it with a structured access framework.

The proposed model would work like this: a vetted list of “trusted partner” countries or companies would be exempt from the export restrictions. Those entities would be allowed to deploy Fable 5 and Mythos 5 under agreed security conditions. The same framework could ultimately extend to models from OpenAI and Google.

No formal announcement was expected from the summit’s technology agenda session on June 17. The discussions are ongoing, and no timeline for a resolution has been confirmed.

Why This Is Happening Now

The June 12 export control directive forced Anthropic to shut down Fable 5 and Mythos 5 for all users worldwide, not just foreign nationals, because the company could not reliably segment access at the API level. That meant enterprises in the EU, UK, Australia, Japan, and every other market lost access to models they had been actively evaluating, many of them within days of the public launch.

European governments and enterprise buyers reacted immediately. The shutdown made viscerally clear what had previously been an abstract concern: reliance on a single US AI vendor means reliance on the US government’s current policy posture. Sovereign AI infrastructure arguments, long popular in policy circles, gained new urgency with actual enterprise workflows going offline.

The G7 intervention follows a broader pattern of allied governments pushing back on US export controls they see as imprecisely calibrated. The AI models debate has similarities to earlier disputes over semiconductor export policy, where allied nations argued that blanket restrictions hurt friends as much as adversaries.

What This Means for Your Business

A trusted partner framework, if it materialises, will come with conditions. Access to restricted models is unlikely to be automatic even for allied-nation businesses. It will probably require certification, compliance reporting, enhanced identity verification, or contracting directly with approved entities. The same friction currently affecting Fable 5 would likely apply to any future high-capability model that triggers export concerns.

The current risk is not fixed. Fable 5 and Mythos 5 remain offline as of today. No restoration timeline exists. Any business that built critical workflows on those models needs operational alternatives now, not after negotiations conclude.

This episode has permanently changed the enterprise AI risk calculation. Before June 12, the idea that a US government directive could unilaterally take your AI vendor’s flagship product offline within hours was theoretical. It is now documented experience. Enterprise AI strategy now requires explicit attention to access resilience: what is your fallback if your primary model becomes unavailable, and how quickly can you switch?

Multi-model architecture is not optional. The businesses least disrupted by the Fable 5 shutdown were those already running across multiple providers. Whether that is Anthropic plus OpenAI, Claude plus GPT-5.5, or a mix of hosted and open-source, the principle is the same. No single vendor dependency in production.

Watch the G7 outcome closely. If a trusted partner framework is agreed, the implementation details will matter enormously: who qualifies, how they apply, what compliance is required, and whether access is granted at the company or country level. Those terms will set the precedent for every future model access dispute.

The Bigger Picture

The G7 AI access debate is the first clear signal that frontier AI has entered the same geopolitical category as semiconductors, rare earths, and other dual-use strategic technologies. Governments are now negotiating over which of their allies can run which AI models. That is a new reality, and it will shape enterprise AI procurement for years.

For businesses outside the US, the lesson is not to stop using US AI platforms. The lesson is to treat AI vendor access as a strategic dependency that needs active management, just like cloud infrastructure, financial counterparties, and supply chain concentration. Identifying your dependencies now, before the next policy event, is how resilient organizations operate.

The Washington negotiations continue. The G7 discussions continue. The models remain offline.

---

Enterprise DNA helps businesses build AI strategies that account for exactly this kind of operational and regulatory risk. If you want to talk through what this means for your organisation, book a session with Sam McKay.