The US government has ordered AI developer Anthropic to block access to its recently released Fable 5 and Mythos 5 models, the Tagesspiegel reports. The order affects all foreign users, including those in the US or working at Anthropic. The company had to cut access for all users at short notice.
The reason for the order is the AI’s ability to detect software vulnerabilities that have remained undiscovered for years. This capability was previously used by US authorities and selected companies to close security gaps. The government fears the technology could be used as a dangerous cyber weapon in the wrong hands. According to Anthropic, authorities believe they have found a way to bypass the software’s built-in restrictions.
Anthropic said it received only partial information from the government and disagrees with the decision. The security measures for Fable 5 had been extensively tested. Other providers’ models, such as OpenAI’s GPT-5.5, have similar capabilities. The move is unprecedented: never before has the US government ordered the blocking of an AI model. Anthropic CEO Dario Amodei had recently spoken in favor of government blocking of potentially dangerous AI software – but based on transparent procedures that, in the company’s view, were not followed in this case.
Source: www.tagesspiegel.de
