Anthropic

US lifts the ban, Fable 5 reopens worldwide from 01/07: what did Anthropic commit to in exchange?

Light bursting through at the end of a dark tunnel — a symbol of access being restored

After nearly 3 weeks of being forced offline worldwide, Claude Fable 5 is officially back: on 30/06/2026, the US Department of Commerce lifted the export controls on Fable 5 and Mythos 5, and from 01/07/2026 Fable 5 reopened to users worldwide on Claude.ai, Claude Platform, Claude Code and Claude Cowork. In exchange, Anthropic committed to a set of new safety measures — from a jailbreak-blocking classifier to a HackerOne bug bounty program and deeper collaboration with the US government. This is the (provisional) conclusion of an unprecedented event we covered in our previous article.

Quick summary

  • 30/06/2026: US Commerce Secretary Howard Lutnick signed a letter to Anthropic co-founder Tom Brown — lifting the export controls after ~2 weeks of review with Anthropic.
  • 01/07/2026: Fable 5 reopened worldwide on Claude.ai, Claude Platform, Claude Code, Claude Cowork.
  • 26/06/2026: Mythos 5 had already returned earlier, for ~100 US companies + federal agencies protecting critical infrastructure.
  • New commitments: a classifier blocking the bypass technique (reported by Amazon) in >99% of cases; a HackerOne program for jailbreaks; the US government gets to test models before release.
  • The lesson: access to cloud AI can be cut off — and given back — by administrative decision; businesses need a proactive plan.

What happened?

A quick recap: on 12/06/2026, the US government issued an export control directive forcing Anthropic to block foreign nationals from using Fable 5 and Mythos 5, after Amazon researchers reported a jailbreak technique that got the models to identify software vulnerabilities and write proof-of-concept exploit code. Unable to classify user nationality in real time, Anthropic took both models offline worldwide within hours.

On 30/06/2026, Commerce Secretary Howard Lutnick sent a letter to Anthropic co-founder Tom Brown announcing the removal of the controls, after roughly two weeks in which the Department and Anthropic reviewed the models and remediation measures together. According to Anthropic, Fable 5 has been back in operation for all users since 01/07/2026.

Cybersecurity specialist monitoring multiple screens in a dark room
The price of reopening: a classifier blocking jailbreaks >99% + a HackerOne bug bounty program. Photo: Tima Miroshnichenko / Pexels

Mythos 5 came back first — but only for a select group

Few people noticed: Mythos 5 — the more tightly controlled model — was actually restored on 26/06/2026, but only for around 100 US companies and federal agencies tasked with protecting critical infrastructure. This "open the valve one notch at a time" approach shows the US government is tiering access to powerful AI by how trusted an organization is — a precedent worth watching.

What did Anthropic commit to in exchange?

According to Anthropic's announcement, the price of reopening is a new package of safety measures:

  • A new safety classifier trained to block the exact bypass technique Amazon reported — Anthropic says it blocks more than 99% of cases, with experts from the Center for AI Standards & Innovation (Department of Commerce) participating in the testing.
  • A HackerOne program so security researchers can report new jailbreaks on Fable 5.
  • Deep collaboration with the US government: pre-release testing, rapid information sharing and joint research; the government gets early access to test future frontier models before launch.
  • Proactively hunting for security issues and reporting abusive behavior when detected.

In other words: the model is back online, but the relationship between frontier AI labs and the US government has entered a phase of significantly tighter oversight.

Why this event matters

This is the first complete cycle of a new precedent: a commercial AI model serving hundreds of millions of people was taken offline by the state, then handed back after the company accepted oversight conditions. Two takeaways:

  • Access to cloud AI is conditional. It can vanish within hours and return weeks later — on the regulator's schedule, not your business's.
  • AI safety standards are being rewritten in real time. Pre-release testing with the government, bug bounties for jailbreaks, purpose-built classifiers — these are likely to become industry-wide practice.
Two business people discussing documents in a dimly lit office
The relationship between AI labs and the US government enters a phase of tighter oversight. Photo: cottonbro studio / Pexels

Lessons for Vietnamese businesses

For Vietnamese businesses, the 3 weeks Fable 5 "disappeared" were a free stress test: which of your processes grind to a halt when a cloud model gets cut off? This episode ended well, but nobody guarantees the next one will — with whichever model, from whichever vendor, for whatever reason.

The pragmatic path: use cloud AI where it fits, but for core data and processes, have a plan for in-house AI running on your own infrastructure — open-source models, data on premises, nothing anyone can switch off remotely. That is the philosophy behind Namtech's private in-house AI platform.

Frequently asked questions

Can I use Fable 5 again?

Yes. According to Anthropic, Fable 5 has been back in operation for users worldwide since 01/07/2026 on Claude.ai, Claude Platform, Claude Code and Claude Cowork.

Why did the US lift the ban?

After about 2 weeks of review between the Commerce Department and Anthropic, the company added a classifier blocking the reported jailbreak technique (blocking >99% of cases, per Anthropic) and committed to deeper oversight collaboration; Secretary Howard Lutnick signed the letter lifting the order on 30/06/2026.

What about Mythos 5?

Mythos 5 was restored earlier (26/06/2026) but only for around 100 US companies and federal agencies protecting critical infrastructure.

Could this happen again?

Nobody can guarantee it won't. The precedent now exists: access to cloud AI models can be revoked by administrative decision. Businesses should have a fallback plan — multiple vendors or on-premise in-house AI for core data and processes.

Don't let your AI depend on an administrative decision

Namtech deploys private in-house AI platforms — open-source models running 100% on your infrastructure, with nothing anyone can switch off remotely.

Book a free consultation

Note: This article is compiled from public sources as of 02/07/2026; information is for reference and may change.

Get started

Start with a free assessment

To define the right package and detailed scope, Namtech offers a short, no-cost assessment.

We reply within 1 business day. No spam, we never share your info.