Anthropic’s Mythos 5 is moving back into the hands of a limited group of organizations, but the broader market remains locked out. The U.S. government has approved a restricted restoration of access for selected American entities it considers trustworthy, while companies outside that circle, including those in India, still face barriers.
The decision matters because Mythos 5 is described as Anthropic’s strongest cybersecurity model. More than 100 companies and institutions in the U.S. are expected to regain access, but only after they are cleared under the government’s selection process.
Who gets access first
According to Anthropic, approval from the U.S. government allows the company to reactivate Mythos 5 for selected organizations that help operate and protect critical infrastructure. The company told Reuters that the restoration process will move quickly.
Sources familiar with the matter said many of the approved organizations are Fortune 500 companies, and several are believed to be part of Anthropic’s Project Glasswing program. That program includes well-known technology companies and research institutions.
| Access Group | Status | Notes |
|---|---|---|
| Selected U.S. organizations | Access restored | Over 100 companies and institutions expected to regain access |
| Organizations outside approval list | Still restricted | Businesses in countries such as India remain blocked |
| Anthropic non-U.S. employees | Access allowed | Can work with the model after licensing changes |
The U.S. Department of Commerce also removed an export licensing requirement when trusted U.S. organizations allow their non-U.S. employees to use Mythos 5. Anthropic’s own employees who are not U.S. citizens will also be able to work with the model.
Security concerns drove the restriction
The rollback follows a broader crackdown ordered weeks earlier by the Trump administration, which told Anthropic to stop access to its two most advanced models, Mythos 5 and Fable 5. The main concern was national security, especially the risk that frontier AI systems could be misused.
U.S. officials have warned that highly capable AI models could be exploited by hostile governments or cybercriminals. Experts have also said that advanced cybersecurity models like Mythos 5 could make sophisticated attacks easier if they fall into the wrong hands.
That risk is considered especially serious in sectors that rely on legacy digital systems and dense interconnections, including banking. In that setting, a model like Mythos 5 is seen not only as a productivity tool but also as a strategic technology.
Department of Commerce officials told Anthropic that the company had made “significant progress” in addressing the risks tied to the model. Officials did not say which additional safeguards had been added.
Earlier this month, Anthropic said it understood the government believed there were ways to bypass or break the protections built into Fable 5. Those protections were designed to prevent the model from being used to identify software vulnerabilities.
Debate over government selection
The selective approach has drawn criticism over transparency and fairness. John Coleman, legislative counsel at the Foundation for Individual Rights and Expression, questioned how the companies were chosen and why others were left out.
Coleman said there was no clarity about who was selected or why, arguing that the arrangement gives the government too much power and raises legal and transparency concerns. OpenAI CEO Sam Altman voiced a similar complaint in a post on X, saying extensive safety testing is not a bad idea but that he does not like the government choosing customers.
The same day, OpenAI said it would delay a wider rollout of GPT-5.6 after a request from the U.S. government. As with Anthropic, only a small group of vetted partners will receive early access.
Anthropic is expected to receive permission to release Fable 5 more broadly in the future, but no timeline has been announced. Mythos 5 and Fable 5 share the same base AI model, yet they serve different purposes, with Fable aimed at general use and Mythos stripping away some protections to support advanced cybersecurity work.
The restrictions are tied to a new executive order signed by President Donald Trump. It creates a voluntary framework requiring AI developers to submit certain frontier models to the U.S. government for review before they are released to trusted partners.
Analysts say the temporary system addresses immediate safety concerns, but it also leaves uncertainty about when advanced AI tools will become available to users and businesses outside the government-selected group.
