Amazon CEO concerned about Anthropic model security

Amazon CEO Andy Jassy may have warned the government about security issues that led to global access restrictions on two Anthropic models. According to The Wall Street Journal, Jassy informed U.S. Treasury Secretary Scott Bessent and other officials that Amazon researchers used the Claude Fable 5 model to extract data that could be used in cyberattacks. This is reported by Techcrunch.com reporting .

Following this warning, the U.S. government imposed export control restrictions on the Fable 5 and Mythos 5 models. An Amazon spokesperson told the WSJ that it is standard practice for governments to consult with the company on potential security risks, but details of such discussions remain confidential.

The Information and Reuters also confirmed that Amazon, a major investor in Anthropic, expressed its concerns regarding the security of these models. Reports indicate that the potential for obtaining sensitive data through AI models was assessed as a serious risk.

According to David Sacks, a former AI advisor to the Donald Trump administration, Anthropic and a trusted U.S. government partner reported a "jailbreak" condition in the model. Sacks claims the government asked Anthropic CEO Dario Amodei to fix the error or shut down the model, but Amodei refused.