US Eases AI Model Restrictions, Granting Limited Access to Anthropic's Latest Technology

The U.S. government has begun to cautiously ease restrictions on the use of cutting-edge artificial intelligence (AI) models. AI startup Anthropic announced on June 27 that it has received permission to offer limited access to its newest AI model, Mythos 5, to a select group of U.S. cybersecurity companies.

This partial rollback follows an earlier directive on June 12, when the U.S. government had instructed Anthropic to halt the provision of its latest models, Mythos 5 and Fable 5. The government cited identified vulnerabilities in existing safety measures that could be exploited maliciously, prompting an immediate restriction on access.

An Anthropic spokesperson confirmed that the company has secured government approval to re-open access to Mythos 5 for a few U.S. cybersecurity firms. The company is reportedly in ongoing discussions with regulatory authorities, aiming to gradually expand the scope of usage and eventually restore public access to Fable 5.

This development occurs amidst a complex relationship between Anthropic and the U.S. government. Previously, Anthropic's refusal to allow its AI technology for mass surveillance and autonomous weapons reportedly led to dissatisfaction from the White House, resulting in the cancellation of a contract with the U.S. Department of Defense. Meanwhile, OpenAI also announced the release of its new GPT-5.6 model, initially available in a limited preview to government-vetted partners. The White House has also mandated a national security review mechanism for advanced AI models, encouraging developers to undergo risk assessments before public release, though specific details on implementation remain pending.