Our Take
Washington has weaponized export controls to settle a domestic feud with Anthropic, but the real damage runs to every allied government that just learned their AI infrastructure can be switched off by an email from Washington.
Why it matters
Anthropic refused to build surveillance and autonomous weapons tools for the Pentagon, so the Trump administration blacklisted it—then claimed the shutdown was about a jailbreak. Europe and Canada watched a core dependency vanish in an afternoon and are now treating American AI as a security threat, not an asset.
Do this week
Security and strategy leads: audit your production Claude dependencies by end of week and model fallback to open-source or non-US alternatives before Q3 procurement locks in.
The shutdown happened in four days, with no warning
On June 9, 2026, Anthropic released Claude Fable 5 and Claude Mythos 5, the company's most capable models to date. Fable 5 was available to the general public; Mythos 5 remained restricted to a controlled access programme called Project Glasswing. Four days later, on June 12 at 5:21 pm ET, Commerce Secretary Howard Lutnick sent a letter to CEO Dario Amodei ordering both models offline for any foreign national, anywhere in the world, effective immediately.
The company said it could not filter users by nationality in real time, so it disabled access for everyone. Anthropic employees born outside the US lost access to their own product.
Lutnick cited national security and a jailbreak method that could bypass Fable 5's safety guardrails. Anthropic countered that the vulnerability was narrow: the ability to review code and identify errors, a capability that exists in OpenAI's GPT-5.5 and other competing models. David Sacks, co-chair of the President's Council of Advisers on Science and Technology, disputed Anthropic's account on X, saying the administration had offered a choice—fix the flaw or pull the models—and Amodei refused.
The Wall Street Journal reported that Amazon CEO Andy Jassy had flagged the vulnerability to Treasury Secretary Scott Bessent, telling him that Amazon researchers had used Fable 5 prompts to obtain information relevant to cyberattacks. Amazon holds a substantial stake in Anthropic. An Amazon spokesperson confirmed it is "not uncommon for governments to seek our counsel on potential security risks" but declined to provide details.
This is the tail end of a deeper fight with the Pentagon
The jailbreak was the stated reason. The actual reason goes back months.
Earlier this year, Anthropic took public positions against mass surveillance and fully autonomous weapons systems. That stance infuriated Pete Hegseth, then Secretary of Defense. President Trump ordered every federal agency to stop using Anthropic's technology and designated the company a "Supply-Chain Risk to National Security," a label ordinarily reserved for foreign adversaries like Huawei. Anthropic filed suit to reverse the blacklisting, warning it could cost the company hundreds of millions in revenue.
So Anthropic is now deemed too dangerous for the US government to use AND too dangerous for anyone else to use. Dean Ball, an AI policy expert who served briefly in the Trump administration, called the order "simply cartoonish," noting that an administration willing to export advanced AI chips to China now wants to ban Britain and Canada from using Anthropic's best models.
Europe and Canada just learned that American AI is a jurisdiction problem, not a product
Outside the US, the reaction sidestepped the jailbreak debate entirely. The concern was structural: a critical tool embedded in research labs, government agencies, and commercial operations worldwide had been switched off by a foreign government in an afternoon.
The European Commission confirmed it is examining the fallout. A spokesperson said highly capable AI models offer benefits including for cyber-defence but that "contingency measures taken in this light should not be discriminatory against partners." European politicians were sharper. French commentary framed the shutdown as proof that "Europe cannot settle for being an open market dependent on technologies designed, funded, and controlled elsewhere." Finnish MEP Aura Salla said Europe "cannot continue to increase its technical potential by relying on access that can be turned off by a foreign government overnight."
Canadian Prime Minister Mark Carney, speaking ahead of the G7 summit, said the episode showed the dangers of overreliance on a limited number of American providers and flagged AI as a major summit topic. Britain's AI and Online Safety Minister Kanishka Narayan said the move should drive deeper investment in the country's own AI industry.
The path back runs through the Commerce Department's Bureau of Industry and Security. Anthropic needs an individually validated licence to restore access to either model. Sacks said plainly: fix the jailbreak, lift the control.
For the governments watching from outside, the technical patch is almost irrelevant. The lesson drawn is that access to frontier AI is now a matter of whose jurisdiction holds the switch. Last week, the answer was Washington's. And many capitals didn't like how that felt.