The US government has relaxed some limits on Anthropic’s advanced AI model Mythos 5, but access remains tightly restricted and largely unavailable outside the United States, including India. Instead of a full release, the authorities have allowed the model to be used only by a select group of vetted American organisations, leaving most global users excluded.
This change follows an earlier directive in which the US administration instructed Anthropic to halt access to its most powerful models, Mythos 5 and Fable 5, over national security concerns. The policy reflects a broader effort by the Donald Trump administration to regulate frontier AI systems that are considered potentially dangerous if misused by hostile states or cybercriminal groups.
Anthropic has stated that it has now received clearance to re-enable Mythos 5 for approved organisations involved in protecting and operating critical infrastructure in the US.
According to the company, the model is being restored quickly for these selected users, while discussions continue with the government to possibly broaden access and eventually reintroduce Fable 5 for wider public use.
Reports suggest that over 100 organisations, including major corporations and members of Anthropic’s Project Glasswing network, are part of the approved group. These include large firms and research institutions.
The US Commerce Department has also removed the requirement for export licences in cases where trusted US organisations allow non-American employees to use the model. However, companies outside the approved list remain barred, meaning countries like India still do not have access.
Officials had earlier expressed concern that highly capable AI systems could be exploited by foreign militaries or intelligence agencies, particularly in nations such as China and Russia. Cybersecurity experts have warned that such tools could significantly increase the sophistication of cyberattacks if widely available.
The Commerce Department said Anthropic had made progress in addressing safety concerns, but did not detail what additional safeguards were implemented.
Some critics have raised concerns about the lack of transparency in how access decisions are made. Others argue that government-controlled distribution of AI models concentrates too much power in official hands.
OpenAI CEO Sam Altman also criticised the approach, saying he supports safety testing but disagrees with governments deciding which customers get access.
The policy change comes alongside similar delays in broader AI releases, with only limited access being granted to trusted partners while wider rollout remains uncertain.
Fable 5 is intended for general public use, while Mythos 5 is designed for more technical cybersecurity applications with fewer safeguards. The current restrictions stem from a recent executive order requiring frontier AI models to be reviewed before release to selected users.
