June 29, 2026. The same week OpenAI placed its new GPT-5.6 models behind a government gate, Anthropic moved through that gate from the other side. On June 26, the US government partially lifted the export-control block it had imposed two weeks earlier, clearing Anthropic to bring its most powerful cybersecurity model, Claude Mythos 5, back to a limited set of vetted organizations. For the businesses we advise, it is another data point in the same trend: access to frontier AI is now shaped as much by policy as by product.
What happened
- The block, then a partial lift. On June 12, a US export-control directive forced Anthropic to disable Mythos 5 and the consumer-facing Fable 5 for all customers. On June 26, Commerce Secretary Howard Lutnick wrote to Anthropic that appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model, citing significant progress in daily talks.
- Who gets it. According to Semafor, the clearance covers more than 100 US institutions, including major companies and government agencies, many of them part of Anthropic's defensive security initiative. A license is no longer required for the named entities and their staff.
- Why it was pulled. Reporting from CNN and others ties the original block to concerns that the models could be jailbroken for misuse and to worries about foreign access to Mythos.
- Fable 5 is still waiting. The government's letter is silent on Fable 5, the weaker model that was briefly the most powerful one available to consumers. Anthropic says it is working to expand Mythos 5 access and make Fable 5 generally available again.
What it means for operators
This is the sequel to a story we wrote two weeks ago about model availability risk, and it confirms the warning. A model that is available today can be pulled tomorrow, and a model that is blocked can be selectively released to a vetted few. If your business depends on one specific frontier model, you are exposed to decisions made far outside your control. The practical defense is the one we recommend for every client: keep your automation model-agnostic, maintain a working fallback provider, and run production workloads on widely available models while reserving any single gated model for experiments. That is how we design AI automation for the agencies and founders we work with, and it is why an AI engineer who builds in provider flexibility is worth far more than one who hard-codes a single vendor.
Frequently Asked Questions
Mythos 5 is Anthropic's most powerful model and its strongest for cybersecurity work. It was pulled offline on June 12, 2026 under a US export-control directive, alongside the consumer-facing Fable 5, and was partially cleared for limited release on June 26.
The June 26 clearance covers more than 100 vetted US institutions, including major companies and government agencies, many tied to Anthropic's defensive security work. It is not generally available to the public or to most businesses, and the government's letter did not clear Fable 5.
Reporting links the original block to concerns that the models could be jailbroken for malicious use and to worries about foreign access to Mythos. The government later said appropriate safeguards were in place to allow limited access to trusted partners.
Treat frontier-model access as something that can change with policy, not just pricing. Keep your AI automation model-agnostic, maintain a fallback provider, and run production on widely available models so a block, delay, or limited release cannot stop your operations.