The U.S. and Europe are discussing a “trusted partner” scheme for access to frontier AI models, days after Washington blocked Anthropic from supplying its most advanced tools to foreign users.
The proposal, discussed by U.S. Commerce Secretary Howard Lutnick with European diplomats on the sidelines of the G7 summit in France, would allow close American allies to test cutting-edge models under privileged access arrangements, according to The Financial Times.
Washington weighs trusted access for AI allies
The discussions follow the Trump administration’s decision to restrict foreign access to Anthropic’s Fable 5 and Mythos 5 models on national security grounds, raising concerns in Europe and Silicon Valley that access to U.S.-built AI could become a geopolitical lever.
G7 leaders are expected to debate the “trusted partner” idea in Evian-les-Bains, with Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman also due to attend.
The plan would give selected U.S. allies access to advanced systems while addressing Washington’s security concerns around models capable of finding software weaknesses.
European Commission tech chief Henna Virkkunen urged the U.S. to avoid “discriminatory” measures against partners such as the EU, saying both sides needed to clarify the security risks and the “best way to really mitigate those risks.”
Fable 5 ban turns AI access into security policy
The dispute escalated after the U.S. government ordered Anthropic to suspend access to Fable 5 and Mythos 5 for foreign nationals, including customers outside the U.S. and non-American users inside the country.
Anthropic said it had taken the models offline more broadly to comply, while arguing there had been a “misunderstanding” and saying it was working to restore access.
The models drew scrutiny for their cybersecurity power, with Anthropic presenting Mythos as capable of finding critical security gaps that could help defenders but also expose exploitable flaws if used by hostile actors.
From Mythos and Fable to GPT-5.5
Following the Anthropic dispute, OpenAI is also expanding controlled access to its own advanced cyber models, with the FT reporting that GPT-5.5, considered similarly powerful, is being made available to the EU cybersecurity agency Enisa and NATO, according to people familiar with the matter.
The U.K. AI Security Institute has separately described GPT-5.5 as one of the strongest models it has tested on cyber tasks.
For Brussels, the episode strengthens the push to reduce dependence on U.S. technology, with Virkkunen warning it is “never good to be too dependent” on one company or third countries in critical technologies such as AI, while the Commission works with providers and seeks to boost European investment.
