White House move against Anthropic sparks AI safety debate

The White House ordered export controls on Anthropic’s newest AI models, Fable 5 and Mythos 5, after concerns emerged about their cybersecurity safeguards. The Trump administration banned foreign access to the models on Friday, prompting Anthropic to suspend broader distribution while it assesses compliance. The decision followed research by Amazon scientists who bypassed Fable 5’s safeguards, revealing vulnerabilities it was designed to conceal. Officials have not publicly detailed the risks but stated the controls stem from serious safety issues. David Sacks, co-chair of the President’s Council of Advisers on Science and Technology, criticized Anthropic for not addressing the vulnerabilities promptly, calling the delay ‘bewildering.’ The administration hopes the company will fix the flaws and restore access quickly. Anthropic disputes the severity of the risks, arguing the concerns are basic and apply to other publicly available models. The company stated the export controls lack transparency and fairness, violating principles for blocking unsafe AI deployments. It warned that similar restrictions across the industry could halt new model releases entirely. Fable 5, released last Tuesday, was accessible to the public with safeguards against dangerous capabilities, while Mythos 5, a more advanced model, was limited to cybersecurity professionals. A group of researchers and executives signed a letter urging the administration to lift the controls, arguing the vulnerabilities are standard for models designed to write secure code. They also warned the move risks undermining U.S. AI leadership without addressing real threats. The dispute highlights broader tensions between advancing AI innovation and ensuring safety, as the Trump administration seeks to establish oversight frameworks for advanced AI systems.