Anthropic and US officials negotiate to restore Fable 5 and Mythos 5 — Lutnick floats a 'trusted partners' framework

RESOLVED — Update (July 1, 2026): Fable 5 is back. The US Commerce Department lifted the export controls on June 30, and Anthropic restored Fable 5 globally on July 1 — across Claude.ai, Claude Code, Cowork, and the API. For Pro/Max/Team/select Enterprise it’s included for up to 50% of weekly usage limits through July 7, then usage credits (the weekly-quota model the leaked code foreshadowed). A new classifier blocks the Amazon-reported bypass in 99%+ of cases, and Anthropic, Amazon, Microsoft, and Google are proposing an industry-wide jailbreak-severity scoring framework. The 18-day saga is over. See the full restoration coverage.

Update (June 27, 2026 — Day 15): the “trusted partners” framework has activated. On June 26, a letter from Commerce Secretary Lutnick cleared Mythos 5 — the restricted, most-powerful model — for ~100+ vetted US companies and federal agencies, scoped to cybersecurity, saying “appropriate safeguards are in place.” This is the first real activation of the framework tracked below. But Fable 5 — the public model — was specifically NOT included and remains entirely offline. So the most-powerful model is back for a closed government-approved list, while the public version stays dark; that asymmetry is likely how Fable 5 itself eventually returns (gated, US-verified-users-first via the July 8 ID system). Confirmed by CNBC, TechCrunch, CNN, NBC News, The Hill.

Update (June 26, 2026 — Day 14): the strongest return signal yet — though still not an official restoration. Developers spotted new strings in Claude Code v2.1.190 (reported by Decrypt and the leaker pankajkumar_dev on X) pointing to Fable 5 coming back, and revealing a business-model change: the old “purchased separately from your plan” language was removed and replaced with “You’ve used your included Fable 5 usage for this week. Continuing on Fable 5 uses usage credits.” The read: when Fable 5 returns, it will be bundled into Pro/Max/Team subscriptions with a weekly quota (then usage credits beyond it), rather than sold as a separate paid add-on. This is leaked-code evidence of preparation, not an Anthropic announcement, and access is not confirmed restored — Anthropic’s official statement page still shows only the June 12 suspension. Separately, Anthropic this week accused Alibaba of the “largest known distillation attack” on Claude — distillation being one of the exact risks Fable 5’s classifiers and the export order were meant to address.

Evening update (June 26): two clarifying developments. (1) Anthropic staff explicitly debunked the “you can access Fable 5 now” rumors, stating the company is serving “exactly 0 traffic to Fable 5” and that reports of access are categorically false — confirming the v2.1.190 strings are preparation, not a live restoration. (2) The first legal challenge has landed: a US litigation-tech company, Legion, filed suit June 23 against the Commerce Department in DC federal court over the ban. Legion is US-based but its software-dev team is in Canada — so the “all foreign nationals, wherever located” directive cut its developers off from a tool its platform depends on. The complaint calls the harm “immediate, irreparable, and existential.” It’s a pointed illustration that the export order’s collateral damage extends far beyond Anthropic — to any US company with non-US engineers that relies on Fable-class models.

Update (June 24, 2026 — Day 12): still offline, but the likely restoration mechanism is now clearer. Anthropic’s updated privacy policy (effective July 8) adds government-ID and facial-geometry verification — widely read as the infrastructure for a US-citizens-only Fable 5 restoration that complies with the export order rather than lifting it. It’s conditional (not a blanket ID mandate), and skeptics argue ID checks alone may not satisfy a “wherever located” directive. No confirmed restoration date.

Update (June 23, 2026 — Day 11): still offline. The “coming days” optimism from June 18 has not yet materialized — Fable 5 and Mythos 5 remain suspended worldwide. Two dated developments worth noting: (1) the free-access window expired June 22 as originally scheduled, so Fable 5 would now require usage credits at $10/$50 per million tokens — but that pricing transition is moot while the model is suspended; (2) reporting points to Anthropic’s updated privacy policy with government-ID verification, effective July 8, as a plausible mechanism for a US-first restoration without fully lifting the directive — though Anthropic has not confirmed that framing, and no official restoration date exists. Opus 4.8 remains the fallback.

Update (June 19, 2026): the signals point toward near-term restoration, but the models are still down. Anthropic’s Managing Director of International, Chris Ciauri, told a press conference in Seoul: “We are very confident that in the coming days, the models will become available again.” New detail on the government’s rationale has also surfaced — per reporting on the Lutnick letter to CEO Dario Amodei, officials feared the models could be used by military or intelligence users in China, Russia, or other countries of concern. Amodei and Lutnick are both attending the G7 in Evian-les-Bains, where talks may continue. Prediction markets price restoration at roughly 57% before July 1 and 75% before mid-July — optimistic, but not certainty. Anthropic’s status page still gives no official ETA.

What’s new since the shutdown

Our original shutdown coverage documented the June 12 directive. The developments since, per the Globe and Mail, Bloomberg, Business Standard, and law-firm analysis (Greenberg Traurig):

• The letter is public: Bloomberg published the Lutnick letter — the order came from Commerce Secretary Howard Lutnick, with input from the Bureau of Industry and Security (BIS)
• The scope, restated: halt export “to destinations worldwide and all foreign nationals, wherever located” — the “wherever located” clause is why Anthropic had to disable the models for everyone
• Negotiations happened: senior Anthropic technical staff met Commerce officials June 15-16 in Washington
• A cyber official was in the room: National Cyber Director Sean Cairncross joined Monday’s working-level meeting
• A framework is on the table: Lutnick is floating “trusted partners” — sanctioned access for vetted entities in allied nations
• Still down: both models remain suspended worldwide as of June 17; no restoration date

Why this matters

Three reads.

1. “Trusted partners” would be a template, not a one-off. If Commerce establishes a sanctioned channel for vetted allied-country entities to access controlled US models, that becomes the reusable mechanism for every future frontier-model export question. This is the export-control architecture for AI capability being built in real time — using Anthropic’s Fable 5 as the test case. The precedent matters far beyond this one model.

2. The negotiation tone suggests resolution, not escalation. A working-level technical meeting with the National Cyber Director present — rather than litigation or public confrontation — reads as both sides looking for a face-saving path. Anthropic has consistently called it a “misunderstanding”; the government floating a framework rather than digging in points the same direction. That’s a more constructive posture than the Pentagon procurement exclusion earlier this year.

3. The IPO clock is still ticking under all this. Anthropic filed its S-1 on June 1. A flagship model offline during the pre-listing window is a disclosed risk; a resolved version with a clear compliance framework is far more manageable for a roadshow than an open-ended suspension. Speed of resolution is now an IPO variable, not just a product one — and it sits alongside the broader government-equity-stake debate shaping the AI-lab listings.

What it means for Claude users

Still: use Opus 4.8. Nothing has changed operationally since the shutdown — Fable 5 and Mythos 5 are unavailable; the rest of the Claude line works normally.

If you’re an enterprise weighing whether to wait: don’t plan around a specific return date. If the “trusted partners” channel materializes, expect a vetting and compliance process to regain Fable 5 access, not an automatic switch-back — and possibly geographic or entity-level eligibility limits.

If you’re outside the US: the “wherever located / all foreign nationals” framing means any restored access may come with citizenship- or country-level conditions. Watch the framework details closely if Fable-class capability matters to your work.

The honest caveats

A framework being “floated” is not a framework in force. Nothing is settled. The plan could change, stall, or be abandoned; “trusted partners” has no published country list, entity criteria, or timeline.

Both models are still down. Despite the constructive signals, the operative fact for users hasn’t changed since June 12.

This is reported, not officially announced. Anthropic’s last formal statement remains the June 12 post; the negotiation details come via Bloomberg and the Globe and Mail. Treat the “trusted partners” specifics as reporting until a primary announcement confirms them.

What it changes for Pick Right readers

The actionable guidance is unchanged — stay on Opus 4.8 — but the outlook improved: there’s now a visible path back rather than an open-ended void. We’ve updated the shutdown article and the Claude review to point here.

This remains the top watch-list item: whether “trusted partners” becomes real, what conditions it carries, and when Fable 5 actually returns. For broader context, see the original shutdown coverage, the Fable 5 launch, the independent benchmark results, the Claude review, the government-equity-stake policy explainer, and our explainer on Executive Order 14409 — where the ‘trusted partners’ framework originates.

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