For two weeks, one of the world’s most capable cybersecurity AI systems effectively disappeared.
On June 12, the US Commerce Department ordered Anthropic to suspend access to both Mythos 5 and Fable 5 for foreign nationals, citing national security authorities. Faced with immediate compliance requirements, Anthropic shut down access globally while it negotiated a path forward with regulators.
Now, part of that shutdown has been reversed.
Mythos 5 Is Back — For Approved US Defenders
The Trump administration has authorized Anthropic to redeploy Mythos 5 to a limited group of trusted organizations that operate and defend critical infrastructure in the United States. According to Semafor, the approved list includes more than 100 companies, government agencies, and strategic partners.
The exemption also extends to non-American employees working within those approved institutions — reversing one of the most operationally disruptive aspects of the original directive.
Anthropic confirmed the development publicly: “Mythos 5, our strongest cybersecurity model, can be redeployed to a set of US organizations that operate and defend critical infrastructure.” The company says it’s moving quickly to restore access for those partners while continuing discussions about broader availability.
Fable 5, meanwhile, remains offline for the general public.
The Consumer Model Is Still Waiting
The government’s latest authorization applies only to Mythos 5.
Fable 5 — the consumer-facing version of the same underlying architecture — hasn’t received approval for restoration. Anthropic says negotiations with regulators are ongoing. That distinction matters more than it might seem: Fable 5 was designed with automatic routing that sent certain cybersecurity, biology, and dual-use requests to less capable systems. Mythos 5 removes some of those restrictions for vetted defenders operating under Project Glasswing. The model now returning to service is the more powerful — and more tightly controlled — version of the two.
The Shutdown Began With a National Security Directive
Commerce Secretary Howard Lutnick’s June 12 order offered little public technical detail beyond national security concerns. Reporting later indicated that officials believed a method existed for bypassing safeguards within Fable 5.
Anthropic publicly challenged that interpretation, arguing the reported vulnerabilities were neither novel nor uniquely dangerous — and could be identified using other publicly available models. It complied with the directive anyway. Disagreement didn’t change the legal obligation.
For the first time, a frontier AI company found itself negotiating not over chips, cloud infrastructure, or data center compute deals — but over whether a model itself could remain online.
Why Mythos 5 Drew Government Attention
When Anthropic introduced Mythos-class systems earlier this year under Project Glasswing, it framed them primarily as defensive tools. The company claimed Mythos Preview autonomously identified thousands of severe software vulnerabilities across browsers, operating systems, and critical infrastructure software — including decades-old flaws in OpenBSD, FreeBSD, and FFmpeg — with minimal human intervention after receiving an initial task prompt.
Those capabilities formed the foundation of the argument that defenders needed access to frontier AI before similar capabilities proliferated elsewhere. That mission is now back online, but only inside a government-approved perimeter.
A New Precedent for Frontier AI
Whether this episode represents the first government-mandated suspension of a publicly available frontier model may be debated. What’s harder to dispute is what it establishes going forward.
The United States has demonstrated a willingness to intervene directly in the deployment of advanced AI when national security concerns surface — even when those concerns are disputed by the company itself. The criteria for trusted access remain opaque. The process appeared to be negotiated in real time.
Policy researchers and digital rights advocates have increasingly warned that AI governance frameworks are struggling to keep pace with frontier capabilities, particularly in cybersecurity, where the same systems that defend critical infrastructure can also create new forms of strategic risk. The Mythos episode underscores how deployment decisions are increasingly being made through ad hoc negotiations rather than long-established regulatory mechanisms.
For now, Anthropic is provisioning approved American defenders once again. The rest of the world — and the consumers who briefly experienced Fable 5 — are still waiting. No timeline has been given for when, or if, that changes.
Related: Anthropic’s NLA Breakthrough: We Can Finally Read AI’s Thoughts—But They Might Be Lying
