US Government Orders Anthropic to Cut Global Access to Its Most Powerful AI Models

The United States government has ordered Anthropic to immediately suspend access to its two most capable AI models — Fable 5 and Mythos 5 — for all foreign nationals and foreign employees, citing national security concerns. The abrupt directive, received by Anthropic at 5:21 PM ET on June 12, has sent shockwaves through the AI industry and triggered urgent diplomatic conversations about the future of global access to frontier AI.

A Narrow Jailbreak With Sweeping Consequences

The trigger for the order was a government demonstration showing that Fable 5's safety guardrails could be bypassed. The specific technique involved asking the model to read a codebase and identify software vulnerabilities — a scenario Anthropic describes as a "narrow, non-universal jailbreak."

Anthropic pushed back on the government's reasoning. In a public statement, the company said it "disagrees that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people," adding that "perfect jailbreak resistance is currently impossible for any AI model provider" and that its own safeguards "exceed those of previously deployed models."

Despite the disagreement, compliance was not optional. "The net effect of this order," Anthropic wrote, "is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance." All other Anthropic models remain available.

Diplomatic Fallout at the G7

The fallout reached the highest levels of international diplomacy within days. At the G7 summit in Evian-les-Bains, US Commerce Secretary Howard Lutnick held discussions with European diplomats about establishing a "trusted partner" scheme — a framework that would give close US allies privileged access to cutting-edge AI models. Leaders formally debated the proposal on Wednesday, with Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman both in attendance.

The European Commission's tech chief Henna Virkkunen pressed Washington to avoid "discriminatory" measures against partners such as the EU. "I think now it's important to discuss and clarify together with the US side what kind of security concerns they have had here, and what is the best way to really mitigate those risks," Virkkunen told the Financial Times.

Meanwhile, OpenAI — whose GPT 5.5 model is considered similarly advanced — is continuing to expand access to European institutions, including NATO and the EU's cybersecurity agency ENISA, according to people familiar with the process.

The Industry Reacts

The US tech industry was unusually vocal in opposing the government's action. The SIIA, a trade association whose members include Apple, Amazon, and Google, issued a rare public statement condemning what it called an "unprecedented action," warning that the "arbitrary use of discretionary authority" over frontier models would "undermine efforts to advance adoption of the American AI stack globally."

The concern is existential for US companies competing internationally: if access to American AI can be switched off at any moment on national security grounds, foreign governments and enterprises have a powerful incentive to build or adopt alternatives — whether from European players like Mistral or Chinese competitors.

A Two-Tier AI World Takes Shape

The "trusted partner" framework, if adopted, would effectively replicate the structure of semiconductor export controls in the AI domain — creating a tiered global market in which access to the most powerful models depends not on commercial terms but on geopolitical standing.

For Anthropic, the immediate consequences are significant. Mythos 5 had been positioned as a breakthrough cybersecurity tool capable of identifying critical software vulnerabilities, and had only recently been extended to select European institutions and companies. That access is now suspended.

The deeper risk is strategic. Treating frontier AI as a controlled export — like advanced chips — may slow adversarial access in the short term, but AI models are software: they can be replicated, reverse-engineered, or independently developed. Critics argue the ban may accelerate the very fragmentation and capability proliferation the US government is trying to prevent, while alienating the allies whose cooperation is essential to any meaningful global AI governance framework.

Anthropic said it is working with the Trump administration to address the concerns and has pledged to share more details as the situation develops.

Sources: Anthropic public statement (June 12, 2026); Financial Times G7 reporting (June 2026)