After two weeks of uncertainty, the Trump administration has authorized Anthropic to restore access to Claude Mythos 5 for more than 100 specific US companies and government agencies — including permitting non-American employees of those organizations to use the model.
It's a significant softening of the June 12 ban that forced Anthropic to pull both Mythos 5 and Fable 5 from the market entirely, following the administration's concern that the models' cybersecurity capabilities could be misused by foreign nationals.
The Timeline
• June 9: Anthropic releases Fable 5 (a publicly available version of Mythos with additional safety guardrails)
• June 12: The US government orders Anthropic to remove access for any foreign national; Anthropic pulls both models entirely
• June 15: TechCrunch reports the ban was never actually about an AI jailbreak — the government was concerned about the models' raw capability in cybersecurity, biology, and chemistry
• June 26: Commerce Secretary Howard Lutnick writes to Anthropic's chief compute officer Tom Brown, authorizing Mythos 5 deployment to trusted US partners
• June 27: The release formally announced
Who Gets Access
According to a letter from Commerce Secretary Howard Lutnick seen by Semafor:
> "I have determined that appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model."
The authorized list includes:
• Over 100 specific US government agencies and companies that operate and defend critical infrastructure
• Their non-American employees — a notable reversal from the original ban that prohibited foreign nationals entirely
• Anthropic's own non-American employees, who were originally included in the foreign-national restriction
Semafor and Reuters both reported on the development, with Anthropic confirming on X (formerly Twitter):
> "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. We're restoring access for these organizations quickly."
What About Fable 5?
Notably, the administration's directive did not address the release of Fable 5, which was the publicly available version of Mythos with what Anthropic described as "more protections." Both models were pulled after security researchers allegedly bypassed those guardrails with relative ease.
Anthropic said it's "continuing to work with the government to expand access to Mythos 5 and make Fable 5 available for general use again" — but there's no commitment on a timeline for Fable 5's return.
Why This Matters
The Mythos 5 situation is the clearest signal yet that the US government is creating a two-tier model access system: one for US-verified entities and another for everyone else.
This follows President Trump's recent executive order on AI oversight, which asks certain AI companies to voluntarily submit their most advanced models for government review up to 30 days before release. Critics argue this has created a "de facto involuntary licensing regime" for frontier AI.
For developers and enterprises outside the US, this is a worrying trend. If you're building on Anthropic's models and you're not a US critical infrastructure organization, access to the most capable models is at the government's discretion. The same dynamic is now playing out with OpenAI's GPT-5.6 rollout, which was also restricted to a government-vetted partner list.
What Mythos 5 Can Actually Do
Mythos 5 was Anthropic's most capable model for cybersecurity work. The key capabilities that triggered government concern:
• Vulnerability research and exploitation: Mythos can identify and chain software vulnerabilities at a level approaching human security researchers
• Long-horizon security tasks: It excels at multi-step hacking scenarios that require planning and adaptation
• Competitive with GPT-5.6 Sol: OpenAI's own benchmarks show GPT-5.6 Sol is "competitive with Mythos Preview" on ExploitBench
The government's concern was never about naive guardrail bypass — it was about the model's inherent capability to accelerate offensive cyber operations, particularly in the hands of state-aligned actors.
Practical Implications
1. If you're a US company operating critical infrastructure, you can likely request access to Mythos 5 through Anthropic's enterprise sales team 2. If you're a non-US developer, Mythos 5 and Fable 5 remain unavailable. Plan your security tooling around available alternatives (GPT-5.6 Terra/Luna, Claude Fable 4, Google Gemini) 3. This precedent matters for the next release cycle. Both OpenAI and Anthropic are now operating under government-negotiated access for their most capable models. Expect this to become the new normal, not an exception 4. Open-weight models become more valuable. Models that don't require API access — and therefore can't be remotely restricted — are increasingly attractive for security research teams that need consistent access
What's Next
Anthropic is pushing for broader access and a return of Fable 5. OpenAI has signaled it will work with the administration on a formal framework for future model releases. The question is whether this framework will be transparent and predictable — or case-by-case and opaque.
For now, if you need access to frontier AI models for security work, build relationships with multiple US partners and have fallback options ready.
Sources
• TechCrunch — "Trump Admin releases Anthropic Mythos to be used by more than 100 US companies, agencies": https://techcrunch.com/2026/06/26/trump-admin-releases-anthropic-mythos-to-be-used-by-more-than-100-us-companies-agencies/
• Reuters — "US allows Anthropic to release Mythos to 'trusted partners'": https://www.reuters.com/technology/us-releases-anthropic-model-mythos-some-us-companies-semafor-reports-2026-06-26/
• Semafor — original report on the directive: https://www.semafor.com/article/06/27/2026/us-releases-powerful-anthropic-model-mythos-to-some-us-companies
• TechCrunch — "The US government's Anthropic models ban was never about an AI jailbreak": https://techcrunch.com/2026/06/15/the-us-governments-anthropic-models-ban-was-never-about-an-ai-jailbreak/
• OpenAI — "Previewing GPT-5.6 Sol" (Mythos comparison): https://openai.com/index/previewing-gpt-5-6-sol/