The Trump administration gave Anthropic just 90 minutes to cut off global access to its latest AI models. The order, issued last Friday by US Commerce Secretary Howard Lutnick, forces the company to get a license before exporting its Mythos and Fable 5 models anywhere in the world — including to non-US employees working inside America.
Anthropic says it got explicit approval from the Commerce Department before releasing Fable 5. The department reviewed the model and either missed the security flaw or decided it wasn't serious. Now the government says Amazon's Andy Jassy flagged a potential "jailbreak" — a way to bypass the safety guardrails built into Fable 5 — and the administration panicked.
Every AI model has potential jailbreaks. OpenAI and Google have similar vulnerabilities in their own models. AI experts say the threat has been overstated. But the Trump administration has had it out for Anthropic since day one. Donald Trump has called the company "radical left" and its executives "leftwing nutjobs." Earlier this year, he ordered all government agencies to stop using Anthropic's tools because the company insisted its models couldn't be used for certain tasks.
The export controls are an existential threat to Anthropic. Like every major AI developer, it relies heavily on foreign talent. The directive says even non-US citizens working inside America can't access the models. That means Anthropic can't use a significant chunk of its workforce to develop or improve its AI.
The bigger picture is even messier. The Trump administration tore up Biden's AI review framework early in the term, at the urging of Silicon Valley billionaires who helped fund Trump's campaign. Last month, it announced a deal with leading AI developers to have new models evaluated 90 days before release — then backtracked, then reinstated it with a 30-day timeline. Now it's using export licensing, backed by criminal penalties against both the company and its executives, to freeze the world's leading AI model.
The message to the rest of the world is clear: don't rely on American AI. If the US can cut off access with 90 minutes' notice, building those models into your infrastructure is a massive risk. China's open-source AI products start looking a lot more attractive.
For Australia, this is a wake-up call. Australian companies and government agencies have been heavy users of US AI models. If the Trump administration can weaponise access against allies — and it has shown it will — then relying on American AI is betting on a partner that can pull the rug at any moment.
The irony is that the administration may have just handed China a gift. By making US AI look unreliable, it's pushing the rest of the world toward Chinese alternatives. And by restricting Anthropic's access to foreign talent, it's potentially down-skilling America's own AI workforce.
Anthropic has said it will comply with the order while exploring legal options. But the damage is already done. The company's ability to compete globally has been severely undermined, and the US's reputation as a reliable AI partner is in tatters.