The limitations of Anthropic AI may begin to dissolve. After being forced offline earlier this month due to US government security concerns, the company’s most advanced AI models are slowly making a comeback. According to a new report from Axios, Anthropic has already returned the Mythos 5 to a limited number of loyal users, while the Fable 5 could return as early as next week if ongoing negotiations with government agencies continue to progress.
Mythos returns first, while Fable waits in the wings
According to Axios, the US Department of Commerce has cleared Anthropic to restore access to Mythos 5, its robust AI model focused on cybersecurity, to a select group of trusted customers. Unlike Myth 5, Mythos has not been widely available to the public and includes additional security tools designed to reduce the risk of misuse in areas such as cyberattacks and biological threats.
The big prize, however, is Fable 5. Axios reports that the Trump administration is close to lifting restrictions on the model after nearly two weeks of negotiations with Anthropic. The model was pulled just a few days after its launch despite quickly gaining a reputation as one of the most efficient AI systems in the industry, especially for coding and deep thinking. Developers who had already integrated Fable 5 into their workflows were left scrambling when access disappeared overnight.
Commerce Secretary Howard Lutnick reportedly told Anthropic that the company has made significant progress in resolving government issues affecting Mythos 5 and Fable 5, while Anthropic has agreed to continue working with US agencies on future release agreements. However, Axios notes that final approval for Fable 5 still depends on additional agencies, including the Pentagon and the National Security Agency.
This is quickly becoming the new normal on the AI frontier
Interestingly, the return of Mythos 5 says as much about AI control as it does about Anthropic. The company initially positioned the Fable 5 as its flagship public model, while the Mythos 5 is a tightly controlled alternative with additional protections for high-risk use cases. Allowing Mythos to return first suggests that regulators are increasingly comfortable approving AI systems that send them down a hard road before allowing their more open counterparts back into the wild.
The big picture is that this is no longer just an Anthropic problem. OpenAI followed a remarkably similar strategy with its recent GPT-5.6 preview, limiting access to trusted partners while working through the same government review process. If there’s a trend from all of this, it’s that introducing the world’s most capable AI models is no longer just an engineering milestone. It has become the law again
