Mythos 5 Is Back — For Some
Anthropic announced Thursday that the US government will let it redeploy Claude Mythos 5 to a limited set of organizations. Specifically, groups that operate and defend the country’s critical infrastructure.
The news, posted on X by Anthropic’s official account, is the first break in the export controls that effectively killed both Mythos 5 and Fable 5 on June 12.
Mythos 5 is Anthropic’s strongest cybersecurity model. It was designed for threat detection, vulnerability analysis, and defensive security operations. The government’s decision to let it back into the hands of critical infrastructure organizations makes sense from a national security standpoint you want your power grids, water systems, and energy networks defended by the best tools available.
But Fable 5 is still locked.
What Anthropic Said
The company’s statement is carefully worded:
Since June 12, we’ve been working closely with the US government to restore access to Claude Mythos 5 and Fable 5. Today, the government notified us that Mythos 5, our strongest cybersecurity model, can be redeployed to a set of US organizations that operate and defend critical infrastructure.
Anthropic said it is restoring access for these organizations quickly and continuing work with the government to expand access to Mythos 5 and make Fable 5 available for general use again.
The Saga So Far
The two-week saga has been the most dramatic test of US AI export control policy since the June 2 executive order on secure deployment of frontier AI models.
On June 12, reports surfaced that the US government had shut down access to Anthropic’s most advanced models. The story grew when it emerged that Amazon CEO Andy Jassy had personally raised concerns about Anthropic’s models to Trump administration officials conversations that triggered the export controls that took both models offline globally.
Since then, every frontier AI lab has been operating under a new reality: the government can and will decide which models are deployable, to whom, and under what conditions.
Earlier today, OpenAI launched GPT-5.6 Sol, Terra, and Luna under a government-approved limited preview. Sam Altman publicly endorsed the arrangement, calling it quite reasonable.
What This Means
This partial Mythos 5 return is significant for three reasons.
First, it proves the government is willing to negotiate. The June 12 shutdown looked absolute. Today’s announcement shows there is a path to restored access, at least for narrowly defined use cases and authorized organizations.
Second, it creates a tiered access model. Mythos 5 is available to critical infrastructure operators but not to the general public, not to researchers, and not to businesses outside that designation. This is the first concrete example of what AI access control looks like in practice: the government decides who gets which model and why.
Third, Fable 5 remains the open question. Fable was Anthropic’s most capable model across general tasks, not just cybersecurity. If the government allowed Mythos back for defensive security work but is still blocking Fable, it suggests they see general capability as the line, not safety.
The Bottom Line
Anthropic got its cybersecurity model back, but only for a designated set of organizations that protect the country’s critical infrastructure. Fable 5 is still locked. GPT-5.6 is shipping under government approval customer by customer.
The model that was supposed to be the most open and accessible is now the most heavily controlled. And on current evidence, that is not going to change anytime soon.
