On 26 June 2026, the US government formally lifted its two-week block on Anthropic's Claude Mythos 5, granting access to more than 100 vetted companies and federal agencies. Commerce Secretary Howard Lutnick confirmed in writing that "appropriate safeguards have been put in place" to allow redistribution among trusted partners. If you are not on that list, you still cannot access the model.
TL;DR
- The US Commerce Department ended the Mythos 5 suspension on 26 June 2026, two weeks after imposing it over national security concerns.
- Access is restricted to 100+ "trusted partners" — largely Fortune 500 companies and federal agencies — through Anthropic's Project Glasswing gated programme.
- Fable 5, the less powerful variant, remains blocked with no confirmed timeline for restoration.
- The original block followed warnings that the model could be jailbroken and reports that a China-linked group had gained access.
- This marks the emergence of a "government-gated launch" regime now applied to both Anthropic and OpenAI.
What Triggered the Original Block on Mythos 5?
On 12 June 2026, the US government issued an export control directive suspending all access to both Mythos 5 and Fable 5. Two concerns drove the action: Amazon flagged that the models could be jailbroken to bypass safety controls, and intelligence reports indicated a China-linked group had already gained access to Mythos 5.
Given what Mythos 5 can do, the urgency was understandable. On the Firefox 147 cybersecurity benchmark, Mythos 5 developed 181 working exploits compared to just 2 for its predecessor. It autonomously discovered thousands of high- and critical-severity vulnerabilities, including a 27-year-old OpenBSD bug, a 17-year-old FreeBSD remote code execution flaw, and a 16-year-old FFmpeg vulnerability. A model that finds decades-old bugs in critical infrastructure is, by definition, dual-use technology.
For broader context on how this fits alongside other frontier model restrictions, see our analysis of the June 2026 AI tech selloff.
How Did Anthropic Get the Block Lifted?
Anthropic engaged in daily negotiations with the Commerce Department over the full two-week suspension period. The company committed to working with the government on protocols and standards for future model releases — effectively accepting a co-regulatory framework rather than contesting the block.
Anthropic confirmed the outcome on X on 27 June: "We have received word from the U.S. government that Mythos 5, our most powerful cybersecurity model, may be redistributed among a small group of cyber defenders and infrastructure providers."
The language matters. "Redistributed among a small group" is not a public release. This is controlled dissemination to pre-approved entities.
Who Gets Access Through Project Glasswing?
Access flows through Project Glasswing, Anthropic's gated access programme. Confirmed launch partners include AWS, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, Linux Foundation, Microsoft, Nvidia, and Palo Alto Networks.
The full list exceeds 100 institutions, spanning Fortune 500 companies and federal agencies. If your organisation is not already in the programme, there is no public application process announced at this time. For a detailed breakdown of how gated access works for both Mythos 5 and Fable 5, see our gated access guide.
What About Fable 5?
Fable 5 — the less powerful variant that was suspended alongside Mythos 5 — remains blocked. Anthropic has confirmed it is in discussions with the government to restore Fable 5 access but has provided no timeline. The asymmetry suggests the government views even reduced-capability cybersecurity models as requiring additional safeguards before broader distribution.
Is This the New Normal for Frontier AI Releases?
Yes, or at least it is heading that way. One day before the Mythos 5 block was lifted, on 25 June 2026, the White House asked OpenAI to limit the release of GPT-5.6 Sol to government-approved partners before any wider rollout, citing similar security concerns.
Two separate frontier AI labs, two different models, the same mechanism: government pre-approval before release. This constitutes a de facto "government-gated model launch" category for AI systems with significant cybersecurity or dual-use capabilities. The precedent is now set. Future models with comparable offensive potential will almost certainly face the same process.
For enterprises building on frontier inference infrastructure, the OpenAI Jalapeno chip programme offers another angle on how the competitive landscape is shifting.
What This Means for Cybersecurity Teams
For the 100+ organisations with access, Mythos 5 represents a genuine step change in defensive capability. A model that autonomously finds vulnerabilities surviving decades of human review is not incremental improvement — it is a new category of tool.
The competitive implication is stark: organisations with Mythos 5 access can audit their infrastructure against a capability set that their peers without access cannot match. This creates an information asymmetry in defensive posture that will widen until either the model becomes more broadly available or competitors produce equivalents.
The limitation is equally clear. Mythos 5's offensive capability is precisely why it was blocked in the first place. Any organisation deploying it must accept ongoing government oversight of how the model is used, and the access could be revoked again if conditions change.
FAQ
Q: When was the Anthropic Mythos 5 government release confirmed? A: Commerce Secretary Howard Lutnick confirmed the release in writing on 26 June 2026, with Anthropic publicly acknowledging it on 27 June.
Q: Can any company apply for Mythos 5 access now? A: No. Access is currently limited to pre-approved trusted partners through Project Glasswing. There is no public application process announced for new entrants.
Q: Is Fable 5 also unblocked? A: No. Fable 5 remains under the export control suspension. Anthropic is negotiating its restoration but has given no timeline.
Q: Why was Mythos 5 blocked in the first place? A: The block was triggered by two factors: Amazon warned the model could be jailbroken, and intelligence indicated a China-linked group had accessed it. The suspension was issued on 12 June 2026 under export control authority.
Q: Does this affect OpenAI's GPT-5.6 Sol release? A: Indirectly, yes. The White House separately asked OpenAI on 25 June to limit GPT-5.6 Sol distribution to government-approved partners, establishing the same gated-release pattern across multiple AI labs.
Q: What makes Mythos 5 so sensitive compared to other AI models? A: Its cybersecurity capability is orders of magnitude beyond prior models. On the Firefox 147 benchmark it produced 181 working exploits versus 2 for its predecessor, and it discovered critical vulnerabilities that went undetected for up to 27 years.
This article was produced with AI assistance. Learn about our editorial process.
Discussion
0 comments