Amazon CEO Flagged Anthropic AI Security Risks Before White House Blocked Models

Amazon CEO Andy Jassy was among several tech executives who raised concerns this week with senior Trump administration officials about security risks in Anthropic’s advanced AI models, according to reporting cited by Reuters and The Wall Street Journal. The story helps explain a rare move by the White House on Friday night, when Anthropic was ordered to shut off access to its newest models worldwide after national security restrictions were imposed.

The U.S. government told Anthropic to block all foreign nationals, inside and outside the United States, from accessing two new models, Fable 5 and Mythos 5. Anthropic said the limits were imposed through export-control procedures and that, rather than comply only partially, it disabled access globally. The company said the government had told it there was a way to bypass the models’ safeguards and use them to find cyber vulnerabilities, while Anthropic argued the bypass revealed only “minor” flaws that other public models can also identify.

According to the Wall Street Journal, Amazon security researchers found a flaw in Fable that could make it generate information useful to hackers in cyberattacks. Instead of warning Anthropic directly, Amazon reportedly went straight to the White House, which contributed to the order. Amazon did not confirm whether it had spoken with government officials, and a company spokesperson said governments sometimes consult Amazon on security risks and that such discussions are not shared.

The dispute comes amid a complicated relationship between Amazon and Anthropic, as the companies compete in AI while Amazon also provides cloud services to train and run Anthropic’s models and is one of its investors ahead of a planned IPO. White House adviser David Sacks said the order was issued “reluctantly” after Anthropic CEO Dario Amodei “refused” to fix the jailbreak issue or take the model down. He said the administration hopes the safety problem will be corrected so the export restriction can be lifted and Fable can return to public use. Some experts criticized the move as too broad, since it also affects allied countries and even Canadian and British Anthropic employees. A U.S. official told The Information the administration is not expected to impose similar restrictions on other AI companies.