Amazon CEO Andy Jassy reportedly raised concerns about the security risks posed by certain Anthropic AI models before the US government moved to restrict access to them.
The development has sparked fresh debate over AI safety, cybersecurity, and the role of major tech firms in shaping government policy.
Amazon CEO reportedly flagged security risks
According to a report by The Wall Street Journal, Jassy informed US Treasury Secretary Scott Bessent and other government officials that Amazon researchers had used Anthropic’s Claude Fable 5 model to obtain information that could potentially be used in cyberattacks.
The concerns reportedly contributed to a government decision to impose export control restrictions on Anthropic’s Fable 5 and Mythos 5 models.
The reported discussions were followed by a government crackdown that resulted in worldwide access being cut off for the two Anthropic models on Friday.
The move highlights growing scrutiny of advanced AI systems as governments around the world grapple with balancing innovation and national security concerns.
Responding to the report, an Amazon spokesperson said it is not unusual for governments to seek the company’s input on potential security risks.
However, the spokesperson added that Amazon does not disclose the details of those discussions.
The company also pointed to an update indicating that Amazon Web Services (AWS) had itself been affected by the model shutdown.
Other reports echo similar concerns
Separate reports from The Information and Reuters also stated that Amazon, one of Anthropic’s largest investors, had communicated concerns regarding the security implications of the company's AI models.
The reports suggest that worries over the models’ capabilities had been circulating among both industry and government stakeholders before the restrictions were imposed.
David Sacks, former AI czar under President Donald Trump and current co-chair of the President’s Council of Advisors on Science and Technology, also commented on the controversy.
According to Sacks, a “highly credible trusted partner” of both Anthropic and the US government discovered a successful jailbreak of the model.
He claimed the administration asked Anthropic CEO Dario Amodei to either fix the vulnerability or withdraw the model from deployment.
“Dario refused,” Sacks alleged in his account of the discussions.
Anthropic pushes back
Anthropic rejected the implication that its models possess uniquely dangerous capabilities.
In a blog post, the company argued that the functions reportedly causing concern are already available in other publicly accessible AI models.
The company maintained that the capabilities in question are not exclusive to its systems and should be viewed within the broader AI landscape.
The episode underscores increasing tensions between AI developers, technology companies, and government regulators as advanced models become more powerful.
With governments paying closer attention to potential misuse of AI tools, the debate over security safeguards, transparency, and export controls is expected to intensify in the months ahead.
