The US government has prohibited AI company Anthropic from continuing to operate its latest model, Claude Mythos Preview. According to the Süddeutsche Zeitung, authorities see a high risk of misuse: the AI is capable of uncovering security vulnerabilities in software that have remained undetected for decades – a dangerous cyber weapon in the wrong hands.
The conflict stems from a dispute over the military use of AI. Anthropic had refused to release its models for autonomous weapons systems and mass surveillance in the US. In response, the Pentagon declared the company a supply chain risk, and Donald Trump ordered agencies to stop doing business with Anthropic. The company took the matter to court.
Anthropic has no plans to make Mythos Preview publicly available. However, according to a media report, a small group of unauthorized users gained access to the model on the very day Anthropic announced a limited testing phase. The group has been using the AI regularly since then, but not for cybersecurity purposes.
Germany’s Federal Office for Information Security (BSI) stated that the AI could lead to “upheavals in how security vulnerabilities are handled.” It said it is in contact with Anthropic. The case illustrates how difficult it has become even for governments to control highly capable AI models.
Source: www.sueddeutsche.de
