Anthropic said on June 30, 2026 that the US Department of Commerce lifted export controls on its Claude Fable 5 and Mythos 5 models. Fable 5 becomes available globally starting July 1, 2026, across Claude Platform, Claude.ai, Claude Code, and Claude Cowork.
Key Takeaways
- Anthropic confirmed the export controls on Fable 5 and Mythos 5 lifted as of June 30, 2026.
- Fable 5 rolls out July 1 across four Claude surfaces, included in Pro, Max, Team, and select Enterprise plans at up to 50% of weekly usage limits through July 7.
- Anthropic’s new safety classifier blocks the jailbreak technique that triggered the pause in over 99% of cases, per independent testing by the Department of Commerce’s CAISI.
- Commerce Secretary Howard Lutnick said his department “worked closely with Anthropic” for two weeks before approving Fable 5’s return.
- Mythos 5 access, restored June 26 for select US organizations, remains narrower and gated through the Glasswing program.
What Happened?
Anthropic’s newsroom said the export controls on Fable 5 and Mythos 5 “have been lifted” as of June 30, 2026, ending a restriction that had frozen deployment of both models. The company’s official X account told followers it had “received notice that the Department of Commerce has lifted export controls” and would “begin restoring access tomorrow.“
Claude Fable 5 will be available again globally tomorrow.
— Anthropic (@AnthropicAI) July 1, 2026
After a series of productive conversations with the US government, we’re redeploying the model with a new set of classifiers to target and block more cybersecurity tasks. In the near term, some routine tasks like coding…
According to CNBC, Commerce Secretary Howard Lutnick tied the reversal to a compressed review. Lutnick said:
That two-week window points back to mid-June, roughly when Mythos 5 access was separately restored for a set of US organizations following government approval on June 26. Fable 5’s clearance follows the same government track just a few days later, part of a broader pattern of AI jailbreak data shaping export-control decisions across frontier models.
Fable 5 rolls out July 1 across Claude Platform, Claude.ai, Claude Code, and Claude Cowork, bundled into Pro, Max, Team, and select Enterprise plans at up to 50% of weekly usage limits through July 7, after which usage draws on credits. Cloud-platform re-enablement across AWS, Google Cloud, and Microsoft Foundry is still to follow, so direct-to-Claude access returns before the enterprise cloud channels catch up.
The Fix Behind the Reversal
Anthropic linked the restoration to an improved safety classifier built after a reported jailbreak method forced the original pause. The classifier targets that specific bypass technique and blocks it in over 99% of cases. Independent testing came from the Department of Commerce’s Center for AI Standards and Innovation (CAISI), the government’s own AI-evaluation arm. That federal sign-off likely explains the two-week turnaround.
Why it matters?
A single jailbreak technique was enough to trigger a federal export freeze on a frontier model, and the fix that unwound it was a targeted classifier update independently verified by CAISI, not a broad capability rollback. That sequence signals frontier-model export policy is now tracking specific exploit patches, not blanket capability bans.
Anthropic also outlined four government-collaboration commitments going forward: pre-release access for government partners to evaluate models before broad release, rapid information sharing on jailbreaks and misuse patterns, dedicated research teams on AI security priorities, and common industry standards for security evaluation across frontier model providers. Beyond that, the company proposed an industry jailbreak severity-scoring framework, weighing capability gain beyond existing tools, breadth of affected techniques, ease of weaponization, and discoverability.
That framework would give regulators a repeatable trigger for a freeze versus a patch.
What’s Next?
Cloud-platform re-enablement across AWS, Google Cloud, and Microsoft Foundry has not shipped yet, so enterprise customers on those channels still wait for parity with the direct Claude surfaces. Mythos 5 stays the narrower case, live only for already-approved US organizations, with Anthropic saying “continued coordination” is underway to expand access through the Glasswing program. Whether that program widens Mythos 5 access beyond government-vetted users is the detail worth watching next.
SQ Magazine’s Takeaway
The speed is the real story: a freeze that normally takes months resolved in two weeks because the fix was quick and independently verified and not a broader capability change. That still hinges on the classifier holding over-99% outside CAISI’s own test set.
Whether other labs adopt the severity-scoring framework matters more than this one reversal. Until then, “over 99%” describes this jailbreak, not a guarantee, and Mythos 5’s selective access is still worth watching.
