AI company Anthropic announced on Friday that it has deactivated its newest artificial intelligence models, referred to as Fable 5 and Mythos 5, in compliance with a directive from the Trump administration to prevent their utilization by foreign individuals. This move represents a significant imposition by the U.S. government to limit access to cutting-edge AI models. Anthropic recently unveiled Fable to the public, which is a scaled-down version of the more advanced Mythos, access to which the company has restricted due to cybersecurity concerns.
In a released statement, Anthropic expressed disagreement with the government’s approach, stating that it received the directive on Friday without specific details regarding national security worries. The company believes that any prevention of unsafe deployments should occur through a transparent and fair process based on technical facts, principles which it claims were not upheld in this instance. Anthropic described the situation as a “misunderstanding” and expressed its intent to restore access to the models at the earliest opportunity.
The Commerce Department did not provide immediate feedback on the matter. This action follows closely after President Trump’s executive order, signed ten days ago, outlining a framework for evaluating national security risks associated with advanced AI systems prior to their public release. Participation by AI developers in this assessment process would be voluntary.
When Mythos was introduced in April, Anthropic highlighted the model’s potential to identify software vulnerabilities on a large scale, which, if misused, could pose significant risks to economies, public safety, and national security. Bank of Canada governor Tiff Macklem mentioned discussions surrounding Mythos during the International Monetary Fund’s spring meetings, emphasizing the need for ongoing preparedness among firms, regulators, and policymakers to navigate the challenges posed by rapidly advancing AI technologies. Macklem stressed the importance of robust cybersecurity measures given the capacity of these technologies to expose and exploit vulnerabilities.