The US government has cleared the way for Anthropic to bring two of its most talked-about AI models back online. The company announced on X that it will start restoring access to Mythos 5 and Fable 5 on July 1, after receiving permission from the Department of Commerce to lift the export restrictions that had been in place for the past few weeks.
The ban had blocked all foreign nationals from accessing the models, even those living in the US and working directly for Anthropic. To stay compliant, Anthropic went further and cut off access entirely, affecting a much wider group of users. That decision drew attention to just how quickly the government can intervene when it has concerns about powerful AI systems.
The story had been moving in stages before this full restoration. Just days earlier, the Commerce Department had already given Anthropic partial clearance to redeploy Mythos 5 to roughly 100 US organizations involved in critical infrastructure, specifically through its Project Glasswing initiative. Mythos is Anthropic’s most advanced cybersecurity-focused model, so restoring access to that group made sense as a first step. But what would happen to Fable, the version of the technology built for broader public use, was still an open question.
That question now has an answer. According to the Financial Times, Commerce Secretary Howard Lutnick wrote to Anthropic confirming that Fable could return after the company agreed to “proactively detect and address security risks associated with the models.” The original restrictions appear to have been triggered by warnings from Amazon and other companies who flagged that both models were vulnerable to jailbreaks and could be misused.
Anthropic had pushed back on those concerns when it first announced the access suspension, defending the safety measures it had already put in place. The company outlined a range of safeguards it said were designed to prevent misuse. That defense clearly wasn’t enough on its own to satisfy the government at the time, but it seems Anthropic’s subsequent commitments were enough to move things forward.
This episode is a sign of where things are heading for AI companies operating at the frontier. A few key dynamics are worth noting:
- Governments are willing to act fast when they believe a powerful AI model poses a security risk, regardless of where users are physically located
- Cybersecurity capabilities in AI are getting special scrutiny, especially models like Mythos that are built specifically for that use case
- Companies may need to make explicit commitments to ongoing security monitoring, not just one-time audits, to keep regulators satisfied
For Anthropic, getting both models back online is a relief, but the broader lesson is clear. As AI models grow more capable, especially in sensitive areas like cybersecurity, government oversight isn’t going away. If anything, this case may set a precedent for how future disputes between AI developers and regulators get handled.
