US Opens AI Mythos 5 Access Again, but Only for 100 Selected Users

Logo Anthropic. Photo: China Daily

JAKARTA - The United States government has begun to loosen restrictions on Anthropic's AI model Mythos 5. However, access is not open freely. Only about 100 trusted organizations are allowed to use the AI model for cybersecurity.

Anadolu Agency, citing information in a letter seen by NBC News, reported that the decision came after US Trade Representative Howard Lutnick told Anthropic that certain users could regain access to Mythos 5.

Anthropic said through X that the government has allowed Mythos 5 to be reused by a number of US organizations that operate and protect critical infrastructure. The company calls Mythos 5 its most powerful AI model for cybersecurity.

The restored access is estimated to include about 100 organizations. They consist of government agencies and private companies, according to a source familiar with the policy.

Mythos 5 is designed to identify digital security gaps and support cyber defense operations. This model is used to help organizations strengthen their system protection.

Two weeks ago, Lutnick used export control authority to ask Anthropic to stop accessing Mythos 5 and its consumer AI model, Fable 5. The reason is that there are concerns about national security.

The move against Anthropic comes just hours after OpenAI said it would release the GPT-5.6 model gradually at the request of the federal government.

OpenAI CEO Sam Altman called the gradual launch "bad news". He said OpenAI originally planned to release the model more widely.

Anthropic is still negotiating with US government officials to restore access to Fable 5. The discussion is said to continue throughout the weekend.

According to NBC News, the Donald Trump administration is drafting a framework to test security risks in advanced AI models. The US government also wants to strengthen cybersecurity protection.

The reopening of Mythos 5 access shows that the US government is beginning to tighten the fence against advanced AI models, especially those that touch on cybersecurity and critical infrastructure.