The Trump administration gave Anthropic 90 minutes to pull its most advanced AI models from the market on Friday, escalating a bitter feud that's been brewing for months.
At 1:15 pm, administration officials called Anthropic and told them to suspend access to their latest systems — Fable 5 and Mythos 5 — for any foreign national, citing an undefined national security concern. Anthropic complied by 5:21 pm, after the Commerce Department imposed export controls that effectively forced the shutdown.
The government said it had discovered a method to 'jailbreak' Fable 5 — bypassing security restrictions meant to stop customers from using the AI for hacking or other harm. But Anthropic pushed back, saying the fears were overblown. The company also noted the directive didn't explain the specific national security concerns behind the order.
This isn't a new fight. The two sides have been sparring for months over how Anthropic's AI could be used by the military and intelligence agencies. The Pentagon already labelled the company a supply chain risk. Defence Secretary Pete Hegseth rubbed it in on Saturday, posting: 'Three months ago, @DeptofWar kicked @AnthropicAI out of our building—forever. Every passing day proves why that was the right move.'
David Sacks, the former AI czar in the administration, accused Anthropic of being reckless with the Fable 5 release. He claimed a 'highly credible trusted partner' — one shared by Anthropic and the government — had flagged the jailbreak issue. But a person briefed on CEO Dario Amodei's conversations disputed that, saying Anthropic was happy to discuss the concerns.
Multiple tech firms, including Amazon, spoke to the White House about the security issues. Several sources said a message from Amazon CEO Andy Jassy was the most influential in raising alarm. But some officials noted that Amazon's own document was misleading — the same concerning capabilities exist in OpenAI's top model, 5.5. Amazon declined to comment on the specifics, saying it's not unusual for governments to seek their counsel on security risks.
Cybersecurity experts questioned the administration's move. Katie Moussouris, CEO of Luta Security, said she had seen the research paper that triggered the action. 'It's not a jailbreak,' she said. 'If national defence is the goal, this is an own goal.'
The dispute is far from over. Commerce Secretary Howard Lutnick spoke to Anthropic on Friday and was set to meet with senior company officials again on Saturday evening. Meanwhile, Anthropic — which confidentially filed for an IPO earlier this month after a funding round valuing it at nearly $1 trillion — faces an uncertain path.
For Australian tech watchers, this is a reminder of how quickly the US can crack down on AI companies when national security is invoked. The 90-minute deadline and the lack of clear explanation have left many asking whether the administration is overreacting or genuinely protecting against a real threat.