AI company Anthropic is starting to restore access to its ‘Mythos 5’ and ‘Fable 5’ models as of today. In mid-June, the US government had ordered that only Americans could access the models – which also applied to Anthropic employees. The company then completely blocked both models to comply with the regulation.
The point of contention was alleged security vulnerabilities: The AI behind ‘Mythos’ is particularly good at detecting software vulnerabilities that have gone unnoticed for years. US authorities and selected companies have used this capability to close security gaps. The concern was that such an AI could become a dangerous cyber weapon in the wrong hands. The ‘Fable 5’ model is based on ‘Mythos’ technology but blocks cybersecurity and biotechnology functions.
Last week, Anthropic had already released ‘Mythos 5’ for selected companies after the US Department of Commerce allowed sharing with ‘trusted partners’ without an export license. In a letter, Commerce Secretary Howard Lutnick noted that Anthropic had agreed, among other things, to actively search for and close security risks in the models. The ban had also caused concern in Europe, which is largely dependent on the AI infrastructure of American companies.
Source: Tagesschau
