What happened with Anthropic, Fable 5, and Mythos 5?

On June 12, 2026, Anthropic suspended access to its two most advanced models — Claude Fable 5 (shipped June 9) and Claude Mythos 5 — for every customer worldwide. The suspension followed an export-control directive sent by US Secretary of Commerce Howard Lutnick to Anthropic CEO Dario Amodei the prior Friday. The letter ordered Anthropic to stop exporting the two models to all destinations and all foreign nationals worldwide, citing national-security risk that capabilities in those models could be misused if obtained by foreign military intelligence. Because real-time nationality enforcement across a global cloud API is impractical, Anthropic chose a universal shutdown to ensure compliance rather than try to gate access by passport. Both models are currently offline. Older Claude models (Opus 4.7, Sonnet 4.5, the older Claude 4.5 line) remain available globally.

Why did the US Commerce Department order the suspension?

Per the Lutnick letter and follow-up Reuters reporting, the order cites concerns that reported jailbreak methods could bypass Fable 5 and Mythos 5 safety controls in ways that could materially assist foreign military-intelligence or WMD-relevant work. The legal authority is the Export Controls Reform Act of 2018 (ECRA), which lets the Commerce Department require export licenses for dual-use technology and notify US companies when an export license is required. Anthropic's public position is that the order is disproportionate — the company says the cited jailbreaks are minor, present in other publicly available models, and that the order rests on a misunderstanding of how its safety stack works. Anthropic is meeting with US officials to try to resolve the dispute and has not committed to an unsuspension timeline.

What should developers and teams use instead while Fable 5 is offline?

For coding: Claude Opus 4.7 and Sonnet 4.5 are still available and were the SWE-bench leaders before Fable 5 shipped on June 9 — they're a step down from Fable 5 but very usable. GPT-5.5 (live on AWS Bedrock GA June 11) is competitive and the obvious next choice if you want a non-Anthropic option. Gemini 3.1 Pro is the third major option. For chat and writing: Opus 4.7 remains strong; GPT-5.5 and Gemini 3.1 Pro are fine substitutes. For agentic workflows (Claude Code, Agent SDK), the SDKs still work against Opus 4.7 — code keeps running, just on a slightly older model. Practical advice: do not rewrite your prompts assuming Fable 5 will be back next week. Plan for at least a multi-week negotiation between Anthropic and Commerce.

Are other AI providers affected by similar export controls?

Not yet publicly. The Lutnick order targeted Anthropic's two most advanced models specifically. OpenAI's GPT-5.5 and Google's Gemini 3.1 Pro remain export-available globally as of June 17, 2026. However, the CNBC reporting and the parallel G7 'trusted partners' discussion suggest the US is opening a broader debate about which frontier models warrant export-license treatment. If the Fable 5 / Mythos 5 precedent sticks, expect Commerce to use the same authority on next-generation models from any US frontier lab. The G7 'trusted partners' plan being discussed at the Evian-les-Bains summit is the diplomatic counterweight: a framework that would let vetted entities in close US allies regain access to controlled models. That plan is in discussion, not in force.

Is this a temporary suspension or could Fable 5 be permanently restricted?

Officially temporary. Practically uncertain. Anthropic met with US officials on June 15-16 to argue the order is disproportionate, and the company's public framing treats the suspension as a near-term compliance step rather than a permanent shutdown. The most likely resolution paths are: (1) Anthropic adds nationality gating at the API layer and Commerce permits re-export with restrictions, (2) the G7 trusted-partners framework lands and creates a sanctioned channel for allied access, or (3) Anthropic upgrades its safety stack to close the cited jailbreaks and Commerce withdraws the order. Permanent restriction is the worst-case scenario, but it would set a precedent that frontier US labs cannot ship globally — a major shift that Anthropic, OpenAI, and Google will fight hard. Realistic expectation: 2-8 week resolution.

Quick Answer

Anthropic Suspended Fable 5 and Mythos 5: What the US Export Control Order Means (June 2026)

Published: June 17, 2026

Anthropic Suspended Fable 5 and Mythos 5: What the US Export Control Order Means (June 2026)

On June 12, 2026, Anthropic took its two most advanced models offline worldwide. The trigger was an export-control letter from US Commerce Secretary Howard Lutnick. The mechanism was the Export Controls Reform Act of 2018. The result is the first time a US frontier AI lab has been ordered to halt global access to specific models on national-security grounds. This page explains what happened, why, and what to do if your stack was running on Fable 5 or Mythos 5.

Last verified: June 17, 2026.

TL;DR

What: Anthropic suspended Fable 5 and Mythos 5 worldwide on June 12, 2026.
Why: US Commerce Department export-control order citing jailbreak risk.
Legal basis: Export Controls Reform Act of 2018 (ECRA), dual-use civilian-tech provision.
Who’s affected: Every Anthropic customer using Fable 5 or Mythos 5, US or foreign.
What’s still up: Claude Opus 4.7, Sonnet 4.5, and the older Claude 4.5 line are unaffected.
Timeline: Anthropic met with US officials June 15-16 to negotiate. No restoration date.

The timeline

Date	Event
June 9, 2026	Claude Fable 5 ships, becomes SWE-bench Pro leader (~69.2%)
Friday, June 12	Lutnick letter to Dario Amodei orders suspension
June 12, 2026	Anthropic suspends Fable 5 and Mythos 5 worldwide
June 15-16, 2026	Anthropic meets with US officials at Commerce
June 17, 2026 (today)	G7 leaders discuss “trusted partners” alternative

The order asks Anthropic to halt export “to destinations worldwide and all foreign nationals, wherever located.” That phrasing is unusual — most export orders target specific countries. The wherever-located clause means the order also covers foreign nationals working at US companies, foreign nationals on US soil, and US-based subsidiaries of foreign companies. Real-time nationality enforcement at the API layer is technically and legally messy, so Anthropic chose the universal-shutdown path.

What’s actually still available

Anthropic kept the older models running. As of June 17, 2026, you can still use:

Claude Opus 4.7 — the previous flagship, ~62% SWE-bench Pro, $15/$75 per MTok
Claude Sonnet 4.5 — the workhorse, faster and cheaper than Opus
Claude Haiku 4.5 — cheap and fast for batch / classification
The full Claude 4.5 line — including the older Opus and Sonnet variants

Claude Code (the TUI), the Agent SDK, the Workbench, and all the SDKs still work — they just don’t have Fable 5 or Mythos 5 as a routing option anymore. If your code did model: 'claude-fable-5' you’ll get an error; swap to claude-opus-4-7 and most code keeps running.

What to switch to if you absolutely need frontier capability

The Fable 5 capability cliff is real. Opus 4.7 is a step down on hard coding tasks. Three serious substitute options as of June 17:

GPT-5.5 — Live on AWS Bedrock GA since June 11. Roughly Fable 5-class on coding (SWE-bench Pro in the mid-60s%). The most obvious like-for-like swap.
Gemini 3.1 Pro — Live since February 2026. Strong on long context (1M tokens) and reasoning. Pricing roughly $7-15 per MTok depending on tier.
Wait for Gemini 3.5 Pro GA — Google’s I/O announcement targets GA in the June 23-30 window. If you can hold a week, this is the highest-capability option in the pipeline.

Avoid the temptation to rewrite all your prompts. Most prompts that worked on Fable 5 will work on GPT-5.5 with minimal changes. The bigger risk is your eval harness — re-run it on whichever substitute you pick before pushing to production.

Why Commerce went after Anthropic specifically

Two factors in the public reporting:

Fable 5 and Mythos 5 are the current frontier. When Commerce evaluated which models warranted export-license treatment, they picked the most recent and most capable. The decision is less “Anthropic is bad” than “if any models warrant this, it’s these.”

Jailbreak evidence was already public. The cited jailbreaks were documented in independent red-team reports between June 9 and June 11. Commerce had a concrete basis to point to. Anthropic’s argument that the jailbreaks are minor and present in other public models is technically correct — it just doesn’t change the Commerce evaluation, which is forward-looking and about precedent as much as actual risk.

The G7 “trusted partners” angle

The Evian-les-Bains G7 summit (June 16-18) is hosting the CEOs of OpenAI, Anthropic, MistralAI, plus reps from Meta, Google DeepMind, and others. Lutnick is floating a “trusted partners” framework: vetted entities in close US-allied countries would be able to regain access to controlled US AI models through a sanctioned channel.

If this lands, it becomes the answer to “I’m a UK or French enterprise that needs Fable 5 — what do I do?” — you apply through your government for trusted-partner status, and access flows through the sanctioned channel. The plan is in active discussion, not in force, and the details (which countries, which entities, what compliance burden) are unclear.

What to do this week

If you were running Fable 5 in production:

Today: swap the model string to Claude Opus 4.7, re-run your eval harness, deploy.
This week: spin up a parallel evaluation on GPT-5.5 (Bedrock) so you have a non-Anthropic fallback ready.
Next week: decide whether to dual-source long-term, or whether to keep Anthropic-only and ride out the suspension.
Don’t: sign new long-term enterprise contracts assuming Fable 5 returns on a predictable timeline.

If you were evaluating Fable 5 but hadn’t deployed:

Continue the evaluation on Opus 4.7 — most of what you’d learn transfers.
Add GPT-5.5 to your eval set in parallel.
Wait for the resolution before committing.

Bottom line

The Fable 5 / Mythos 5 suspension is the most consequential US government action on AI in 2026. It establishes that Commerce will use ECRA on frontier models, it forced Anthropic into a universal-shutdown posture, and it opens a multi-month negotiation that will reshape how US labs ship globally. The G7 trusted-partners discussion is the diplomatic counterweight, but the framework doesn’t exist yet.

For developers: swap to Opus 4.7 or GPT-5.5 today, dual-source your stack, and don’t bet on a fast restoration timeline.

Anthropic Suspended Fable 5 and Mythos 5: What the US Export Control Order Means (June 2026)

TL;DR

The timeline

What’s actually still available

What to switch to if you absolutely need frontier capability

Why Commerce went after Anthropic specifically

The G7 “trusted partners” angle

What to do this week

Bottom line

Related on andrew.ooo