How an Amazon warning likely triggered US crackdown on Anthropic’s AI model

The U.S. government intervened to remove Anthropic’s AI model, Fable 5, just days after its launch due to safety concerns. According to the Wall Street Journal, senior administration officials spent nearly 24 hours urging Anthropic to voluntarily withdraw the model after Amazon researchers identified ways to bypass its safeguards. Amazon reportedly found prompts that allowed the model to provide restricted cyberattack-related information, raising alarms about potential security risks. Amazon’s concerns escalated when the Trump administration sought feedback on the model’s safety. The company confirmed it had engaged with government officials but declined to disclose specifics, stating it does not share details of such discussions. Meanwhile, the White House suspected a group linked to China may have already exploited the same bypass method, though Anthropic denied these claims, stating it blocks access from China and that officials never raised the issue with them. Anthropic’s CEO, Dario Amodei, argued that the bypass was limited and did not represent a universal jailbreak. The company was given only 90 minutes to remove Fable 5, with no prior warning of a national security threat. In a blog post, Anthropic emphasized that no universal jailbreak had been found, reinforcing its claim that the model’s safeguards remained largely intact. Despite negotiations, the two sides failed to reach an agreement. By Friday evening, the U.S. Commerce Department imposed export controls on both Fable 5 and its underlying system, Mythos 5, effectively blocking their distribution. The move underscored growing tensions between AI innovation and regulatory oversight in the U.S.