U.S. Bans Foreign Access to Anthropic’s Most Powerful AI Models After Amazon Security Warning

TLDR

• Amazon CEO Andy Jassy raised security concerns with Trump administration officials after researchers found Anthropic’s Fable 5 model could aid cyberattacks
• The White House ordered Anthropic to fix vulnerabilities or restrict the model; Trump approved a full foreign access ban
• Anthropic shut down its Fable 5 and Mythos models for all users to comply with the export restrictions
• Anthropic called the vulnerabilities “relatively basic” and said similar capabilities exist in other public AI models
• The ban could hurt Anthropic’s upcoming IPO and benefit rivals like OpenAI

Amazon CEO Andy Jassy spoke directly with U.S. officials, including Treasury Secretary Scott Bessent, about security concerns with Anthropic’s Fable 5 model. Amazon researchers had used specific prompts to get the model to return information that could help carry out cyberattacks.

That information was supposed to be blocked by the model’s safety guardrails. The findings were shared with White House and security officials, who then held meetings to decide how to respond.

Officials asked Anthropic to either fix the problems or take the model offline. When Anthropic’s CEO Dario Amodei spoke with administration officials on Friday, some felt the company was unwilling to work with government security experts on a fix.

Amazon.com, Inc., AMZN

The administration concluded that blocking foreign access was the most direct way to reduce the risk. President Trump approved the move, despite personal concerns it could slow AI innovation.

Anthropic Shuts Down Top Models

Following the order, Anthropic disabled access to both Fable 5 and its Mythos model for all users — not just foreign ones — to ensure compliance with the new export controls.

The company said the vulnerabilities were relatively basic. Anthropic pointed out that other publicly available AI models can surface the same information.

Cybersecurity researcher Andrew Morris, founder of GreyNoise Intelligence, reviewed Amazon’s findings. He said while Fable 5 could identify software bugs in at least four programs, there was no evidence researchers accessed the more dangerous capability of converting those bugs into working exploit code.

Anthropic has long emphasized safety as a core value. The company previously delayed broader access to Mythos at the White House’s request and works with a government AI testing unit before releasing new models.

What This Means for Anthropic

The timing is difficult for Anthropic. The company is preparing for a potential IPO as soon as this fall.

With its top models offline, customers may turn to competitors. OpenAI has its own cyber-focused model and has been in active discussions with the Trump administration.

Tensions between Anthropic and the White House are not new. The Pentagon previously designated Anthropic a security risk, a move the company is fighting in two separate lawsuits.

National Cyber Director Sean Cairncross and Commerce Secretary Howard Lutnick were involved in the discussions leading to the ban. The Commerce Department oversees export controls on critical technology.

Commerce’s ban currently prevents foreign governments, companies, and individuals from accessing Fable and Mythos. Many of Anthropic’s own researchers are foreign-born, which the company said effectively stops them from working on the models.

White House AI adviser David Sacks said the restriction was issued “reluctantly” and expressed hope that Anthropic fixes the issue so the models can return to general release.

The post U.S. Bans Foreign Access to Anthropic’s Most Powerful AI Models After Amazon Security Warning appeared first on CoinCentral.