US government clears Anthropic to release Mythos 5 to ~100 'trusted partners' — but Fable 5 stays offline

What happened

Per CNBC, TechCrunch, CNN Business, NBC News, and The Hill:

• Date: June 26, 2026 — a letter from Commerce Secretary Howard Lutnick to Anthropic.
• What was cleared: Claude Mythos 5 — the restricted, highest-capability model — for roughly 100+ US companies and federal agencies.
• Scope: cybersecurity-oriented applications; the full partner list was not published.
• Lutnick’s language: “appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model,” and Anthropic’s risk-mitigation work since June 12 has “yielded significant progress.”
• What is NOT covered: Fable 5, the publicly available model. It remains entirely offline, 15 days into the suspension, restoration still described as a hope, not a date.
• The arc: this partially ends the two-week standoff that began with the June 12 export-control shutdown.

Why this matters

Three reads.

1. The “government-gated AI” model is now real, not theoretical. For two weeks we’ve covered the idea of “trusted partners” — first as language in Executive Order 14409, then as something Lutnick floated in negotiations. June 26 turns it into operational policy: the US government now decides, entity by entity, who may access a frontier model, and it just approved ~100 of them for Mythos 5. This is a genuine regime change in how the most capable AI is distributed — closer to how the government controls advanced chips or defense technology than how software has historically shipped. Whatever you think of it, it’s the new baseline.

2. The asymmetry is the story: capability returns to the vetted, not the public. The model that came back is the restricted, most powerful one — to a closed list. The model still dark is the public one. That inverts the usual order, where the masses get a capped version and specialists get more. Under export-control logic it makes sense (a closed list is auditable; the open public is not), but the implication for ordinary users is stark: the frontier can be available to approved cybersecurity firms and federal agencies while being entirely unavailable to you. Frontier capability and public availability have been decoupled.

3. It validates the negotiation path — and previews how Fable 5 likely returns. Anthropic spent two weeks meeting Commerce officials and building compliance infrastructure (including the government-ID verification system effective July 8). The Mythos 5 clearance shows that work paying off — “significant progress,” in Lutnick’s words. It also signals the template for Fable 5’s eventual return: not a blanket switch-back, but a gated re-release, most likely US-verified-users-first once the ID system is live. The leaked Claude Code v2.1.190 strings pointing to a weekly-quota Fable 5 fit that picture.

What it means for AI-tools buyers

For almost everyone: nothing changes today. You can’t access Mythos 5 — it’s a government-approved partner program, not public or general-developer access. And Fable 5, the model you’d actually use, is still offline. Claude Opus 4.8 remains the recommended default, and the Claude review stands.

If you’re at a cybersecurity firm or federal agency: you might be on the list — but the list isn’t public, so access comes through Anthropic’s enterprise/government channels under the safeguards, not a self-serve signup. If frontier cyber-capability matters to your org, that’s the door now.

For everyone tracking the bigger picture: this is the clearest signal yet that US frontier AI is being treated as strategic, government-supervised technology. The practical takeaway for planning: assume the most capable models may reach approved institutions before the public, and that public access to top-tier capability can be switched off and conditionally switched back on. Build with that volatility in mind — keep a non-frontier fallback (Opus 4.8, GPT, Gemini) for anything mission-critical.

The honest caveats

Fable 5 is still fully offline. This is the easiest fact to lose in a “Mythos returns” headline. The model the public cares about is not back, has no confirmed date, and was specifically excluded from the June 26 clearance. Day 15 and counting.

The partner list is secret. “~100+ companies and agencies” is the reported figure; the actual list isn’t published. We can’t independently verify who’s in, and “trusted partner” is a designation the government controls without public criteria so far.

“Significant progress” is the government’s framing. The letter is a policy statement from one side of a negotiation. What safeguards Anthropic actually implemented, and how they were judged sufficient, isn’t fully public. Treat the characterization as official, not independently audited.

Government-gated access raises questions this article doesn’t settle. Who qualifies as trusted, how the list is governed, and whether public access ever fully returns are open and consequential. The mechanism is now real; its accountability is still being written.

What it changes for Pick Right readers

Practically: keep using Opus 4.8 or your current model — neither Mythos 5 nor Fable 5 is available to you. Strategically: the June 26 clearance is the moment the “trusted partners” framework stopped being a proposal and became how the US releases frontier AI. The most capable models can now come back to a vetted list while the public version stays dark — and that asymmetry is likely how Fable 5 itself eventually returns, gated behind the ID verification Anthropic switches on July 8.

For the full thread, see the Fable 5 / Mythos 5 government shutdown, the restoration-and-trusted-partners tracker, the Executive Order 14409 explainer, the ID-verification coverage, the Fable 5 launch, the Claude review, and the Alibaba distillation accusation for the parallel cyber-security thread.

Questions or corrections? Email Pick Right. Want the full list? See all news.