Emmanuel Macron spent Wednesday in Évian-les-Bains trying to engineer a workaround to a U.S. policy that, five days earlier, had stripped every non-American user off Anthropic’s two most capable models. The French president convened G7 counterparts alongside Anthropic’s Dario Amodei and OpenAI’s Sam Altman, with bilateral sessions slated for Meta and OpenAI executives, according to a diplomat familiar with his thinking.
The mechanism under discussion is a “trusted partners” framework, raised principally with Commerce Secretary Howard Lutnick at Monday’s opening summit dinner, per two diplomatic sources. The partners could be countries, or companies, or some hybrid arrangement that lets allied governments and firms regain access to Fable 5 and Mythos 5 without formally undoing the export order.
That order is what created the diplomatic emergency. On Friday, Anthropic disabled access for all users after President Trump directed the company to block foreign nationals on national security grounds. Mythos had been deployed in more than 15 countries before the ban, much of it inside critical infrastructure. Allies woke up cut off.
The legal scaffolding is already wobbling. Alasdair Phillips-Robins, a former senior Commerce Department policy adviser, wrote on X that Lutnick’s letter is “so badly drafted it might not restrict API/chatbot access at all.”
The domestic logic is also fraying. Federal officials including Treasury Secretary Scott Bessent learned last week from Amazon chief Andy Jassy that internal research had used Fable 5 to identify flaws in Amazon’s own cybersecurity software, an awkward data point for a ban premised on the model’s offensive potential.
Trump didn’t raise the export ban at Wednesday’s working lunch, a person familiar with the meeting said. Afterward, he told reporters that negotiations with Anthropic over Fable and Mythos were “going fine.” Further guidance on a partners framework is expected in the coming weeks.
The pattern is familiar from Cold War dual-use export regimes: an instrument designed to deny adversaries promptly becomes the thing allies have to negotiate around.
Sources
- Macron Seeks Way Around Trump’s Ban on Anthropic’s AI Models, Bloomberg
- Trump’s Anthropic Restrictions Rattle U.S. Allies as AI Leaders Gather at G-7, Washington Post
- Trump and world leaders joined by OpenAI, Anthropic, Google at G7, CNBC
- G7 leaders discuss ‘trusted partners’ access to cutting-edge U.S. AI models, Japan Times
- Feds’ Legal Basis for Ban on Anthropic’s Most Powerful Models Looks Increasingly Shaky, Gizmodo