The U.S. government has completely banned access from abroad to Anthropic's top-tier artificial intelligence (AI) models, including "Mythos."
On the 12th (local time), Anthropic said, "The U.S. government, following guidance from national security authorities, has issued export control guidance blocking all foreign nationals from accessing 'Fable 5' and 'Mythos 5.'"
The measure prevents access to Mythos from overseas. Foreign nationals staying in the United States are also barred from accessing the model. As a result, even foreign employees affiliated with Anthropic can no longer use Mythos.
Anthropic said, "To comply with the regulation, we immediately halted the Fable 5 and Mythos 5 services that had been offered to all customers."
Anthropic said it received the guidance from the government at 5:21 p.m. (local time) that day. However, it explained that the letter did not specify details about the national security concerns.
However, Anthropic said, "The government appears to believe there is a so-called 'jailbreaking' method to bypass our safeguards that prevent hacking and the creation of biochemical weapons," adding, "We judge this to be a misunderstanding."
Anthropic said the jailbreaking methods the government identified are widely used in other models, including OpenAI's GPT-5.5.
It added, "We believe the government should have the authority, through a transparent, fair, and technically grounded legal process, to block the deployment of unsafe AI," while noting, "This action does not align with those principles."
U.S. online outlet Axios reported the same day, citing government sources, "The Commerce Department became concerned about security risks after a corporations claimed to have breached Mythos's security walls, and decided to take this action."
The source said, "The government earlier tried to temporarily halt Anthropic's model launch," adding, "When it failed to do so, it issued this export control notice." The source also said the access restrictions will continue for weeks until the U.S. government's security posture is strengthened.
Mythos has raised fears across the global security industry, dubbed the "Mythos shock," after it became known to possess expert-level capabilities in detecting security vulnerabilities.
Anthropic disclosed the development of Mythos in April. However, due to the risks, it had shared it only with some institutions, including the government, technology corporations, and financial companies, for a time. It then released the safety-hardened model Fable 5 to the public on the 9th.