Amazon CEO's warnings trigger export controls on Anthropic models

Business rivalries are messy, but having your own lead investor point the federal government at your latest flagship product is an entirely different level of tech drama.

Just days after Anthropic rolled out its powerful new Claude Fable 5 and Mythos 5 models, the Trump administration slapped the company with an abrupt export control order restricting all foreign national access. Instead of attempting the impossible task of filtering users by nationality overnight, Anthropic pulled the plug on both models worldwide.

Now, reports from The Wall Street Journal and Reuters reveal who lit the fuse: Amazon CEO Andy Jassy.

A friendly heads-up to the White House

Earlier this week, Andy Jassy reportedly met with senior administration officials, including Treasury Secretary Scott Bessent, to raise urgent security alarms. Amazon researchers claimed they found prompt jailbreaks that forced Claude Fable 5 to generate actionable cyberattack information.

David Sacks, co-chair of the President's Council of Advisors on Science and Technology, confirmed that the government acted on information from a trusted partner. According to Sacks, the administration demanded Anthropic CEO Dario Amodei patch the exploit or decommission the model immediately. Amodei refused, triggering the June 12 export directive.

Anthropic pushed back hard. While complying with the order, the company released a statement arguing that pulling a commercial model used by hundreds of millions over a narrow potential jailbreak was completely unjustified, especially when similar capabilities already exist in rival models.

An Amazon spokesperson did not deny the talks, noting only that consulting with the government on emerging security threats is routine practice, even as Amazon's own AWS platform took a direct hit from the sudden takedown.

Why it matters

We think this mess signals a dark, chaotic shift in how frontier AI will be regulated. Forget formal rulemaking and public comment periods; Big Tech executives now have a direct batphone to the White House to turn isolated red-teaming finds into federal shutdown orders against competitors or unruly portfolio companies.

For enterprise teams and developers building on Claude, this should set off immediate alarm bells. If access to flagship models can evaporate in a single afternoon over backroom security claims, multi-model redundancy isn't just an engineering luxury anymore, it is essential insurance.

The precedent here looks deeply flawed. Singling out Anthropic for cyber jailbreaks that plague virtually every frontier model creates a culture of selective enforcement. Instead of incentivizing responsible disclosure, it teaches AI labs that sharing safety findings with investors might just get your servers unplugged.