US Clears Anthropic Mythos 5 for 100 Trusted Partners

US Commerce Secretary Howard Lutnick gave Anthropic the green light on Friday to release Claude Mythos 5 to a restricted list of roughly 100 American companies and federal agencies - ending 15 days of total global blackout on the company's two most powerful models. Claude Fable 5 stays blocked, with no timeline for its return.

Key Status

Item	Status
Claude Mythos 5	Cleared for ~100 US trusted partners
Claude Fable 5	Blocked - discussions ongoing
Original launch	June 9, 2026
Export ban imposed	June 12, 2026
Partial clearance	June 26, 2026
Authority	Commerce Secretary Howard Lutnick
Access basis	"Annex A" entities + their foreign national employees

The Clearance Letter

Lutnick's letter to Anthropic, dated June 26, states that "I have determined that appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model." It grants access to entities named in a classified Annex A attachment - the full list hasn't been made public.

The Commerce Department put a terse spin on the speed of the reversal. A spokesman said: "In just two weeks, we have worked diligently to ensure America remains the global leader in AI while safeguarding our security."

Who Gets Access

The cleared entities are described as US companies and federal agencies - critical infrastructure operators and defenders are specifically mentioned. Based on the Project Glasswing structure that predates the ban, the group likely includes organizations like Apple, Google, Cisco, NVIDIA, Microsoft, Amazon Web Services, CrowdStrike, and JPMorganChase, which were Glasswing founding partners. As of early June, approximately 200 firms had access through Glasswing.

The letter also explicitly covers foreign national employees working at cleared entities, and Anthropic's own non-American staff - both groups lost access when the ban took effect and have now regained it.

What the Letter Does Not Cover

Fable 5, Mythos's public-facing sibling, is not mentioned. A Commerce Department spokeswoman suggested that discussions about Fable's release were "ongoing" over the weekend, with no timeline. Anthropic posted on X: "we're continuing to work with the government to expand access to Mythos 5 and make Fable 5 available for general use again."

The letter is also silent on specific technical safeguards Anthropic was required to implement. The phrase "appropriate safeguards are in place" appears without definition.

Howard Lutnick, US Commerce Secretary, who signed the letter clearing Mythos 5 for trusted partners Commerce Secretary Howard Lutnick signed the letter clearing Mythos 5 for access by roughly 100 US institutions. Source: wikimedia.org

The Paradox: Why the Less Restricted Model Got Cleared First

The most counterintuitive aspect of Friday's clearance is which model was approved. Mythos 5 is technically the more dangerous version - it removes safety classifiers present in Fable 5. To understand why the government cleared it first, you have to understand how the two models work.

Fable 5 is built for public deployment. It includes three classifier-based safeguards: cybersecurity queries fall back to Claude Opus 4.8, biology and chemistry requests reroute to Opus 4.8, and distillation attempts are blocked. These filters trigger in less than 5% of sessions under normal use.

Mythos 5 removes those cyber classifiers for approved partners and offers separate biology access programs for vetted researchers. It's the raw version of the same underlying model.

Fable's Jailbreak Problem

The government's reluctance on Fable 5 traces directly to its classifier-based safeguards being bypassed. As documented in our reporting on the Fable 5 jailbreak, security researchers demonstrated that Fable's fallback triggers could be avoided through prompt manipulation shortly after launch. Anthropic itself acknowledged the limitation: "perfect jailbreak resistance is not currently possible for any model provider."

When safety filters on a public model fail, anyone with API access can reach capabilities that were supposed to be gated. That is a fundamentally different risk profile from a model that is structurally unavailable to unauthorized parties.

Access Restriction as the Safeguard

For Mythos 5, the access restriction IS the safety mechanism. Cleared partners in Annex A have been vetted by the US government. They have accountability relationships with both Anthropic and federal agencies. If Mythos 5 is misused by a cleared entity, there's a paper trail and a legal framework. That's not true of a public API accessed by anonymous developers.

The government's logic, in engineering terms: a system with no guardrails accessed only by authenticated, accountable users is safer than a system with fallible guardrails accessed by everyone. That framing makes the clearance sequencing less paradoxical than it looks.

An official rubber stamp on a document, representing government clearance The Commerce Department letter represents the first formal clearance for any AI model under the new export control framework. Source: unsplash.com

Fifteen Days From Launch to Blackout to Partial Return

The speed of this entire sequence is remarkable. Three weeks ago, Mythos 5 and Fable 5 didn't exist as products. They launched on June 9. Three days later, they were gone.

June 9, 2026 - Anthropic launches Claude Fable 5 and Claude Mythos 5 at $10/$50 per million tokens. Fable 5 goes public globally; Mythos 5 is restricted to Project Glasswing partners and biology researchers. Both are described as the company's most capable models ever shipped.
June 12, 2026 - The Trump administration issues an export control directive requiring government permission before Anthropic can export either model to any destination worldwide, or make them available to any foreign national regardless of location. Anthropic disables both models for all users and posts a statement.
Mid-June 2026 - Lutnick's original letter to Anthropic CEO Dario Amodei becomes public - Amazon CEO Andy Jassy's call with the White House was a triggering factor. The letter threatens criminal and civil penalties for non-compliance and requires Anthropic to treat both models as controlled items under export administration regulations.
June 26, 2026 - Commerce Secretary Lutnick signs a new letter clearing Mythos 5 for entities in Annex A. Fable 5 remains blocked. Anthropic begins provisioning access for the approved set.

What Mythos 5 Can Do for Cleared Partners

For organizations that are back online with Mythos 5, the restored capabilities are sizable. Based on Anthropic's June 9 launch documentation, Mythos 5 offers:

Cybersecurity work without Opus 4.8 fallbacks - critical for the infrastructure defenders likely to be on the Annex A list
Protein design workflows reported to run roughly 10x faster than Opus-class models
Novel scientific hypotheses in molecular biology, preferred over Opus-class models in 80% of evaluations
Autonomous genomics work spanning millions of cells across 138 species

The full benchmark picture for Mythos 5 is covered in the model card. On software engineering benchmarks, it sits above the Mythos Preview numbers disclosed in April - itself already the highest published SWE-bench Verified score at 93.9%.

What Fable 5 Needs to Come Back

The Fable 5 situation is more complicated because the government's concern isn't access control - it's that the model's own safety mechanisms do not hold under adversarial conditions.

Getting Fable 5 cleared likely requires Anthropic to demonstrate one of two things: either significantly hardened classifiers that resist the known bypass techniques, or a technical approach that doesn't rely on classifiers at all. Anthropic has given no public timeline and hasn't described what changes would satisfy the administration.

The commercial stakes are significant. Fable 5 was included free on Pro, Max, Team, and seat-based Enterprise plans from June 9 through June 22. Since June 23, it is priced at $10/$50 per million tokens. Every day it remains blocked is revenue Anthropic can't collect from its largest public model.

What To Watch

Fable 5 clearance timeline. Discussions are described as ongoing. Watch for any Anthropic or Commerce Department statement indicating what technical bar Fable 5 needs to clear.

The Annex A list. The specific companies and agencies cleared for Mythos 5 haven't been published. If that list becomes public, it'll show who the US government considers trustworthy enough to operate a model with frontier cybersecurity capabilities.

A formal framework for frontier AI exports. The June 2 White House executive order on AI innovation and security tasked the NSA with establishing benchmarking protocols for "covered frontier models." This episode suggests those protocols are being developed reactively, with events. A durable framework would define clearance criteria in advance rather than through one-off letter exchanges.

Other labs, other models. No other lab has faced this kind of export control action, but the precedent now exists. The government has demonstrated it's willing to use Commerce Department authority to halt a commercially shipping AI model on national security grounds. Other frontier model launches - especially those with strong coding or cybersecurity benchmarks - may face pre-launch review requirements from now on.

Sources: