Anthropic’s Claude Fable 5 was banned, now it’s back live

Imagine launching the most capable AI model your company has ever built. Then, three days later, your own government switches it off. That’s not a hypothetical. It’s exactly what happened to Anthropic.

The shutdown

Claude Fable 5 and Claude Mythos 5 went live on 9 June 2026. They were Anthropic’s first public release in its new Mythos tier, a step above Opus 4.8. Then, on the evening of 12 June, US Commerce Secretary Howard Lutnick sent a letter to Anthropic CEO Dario Amodei. The order was simple: cut off both models for any foreign national, anywhere, including Anthropic’s own staff. Anthropic reportedly had just 90 minutes to comply. There was no way to check every user’s nationality in time. So the company just switched both models off. For everyone. Worldwide.

What actually triggered it

The government cited national security. But the real story traces back to Amazon. Amazon researchers found a way to prompt Fable 5 into spotting software bugs. In one case, the model even wrote code showing how to exploit one. Amazon flagged this to the White House, and that fed into the shutdown order. Anthropic pushed back once it reviewed the report. It tested the same prompts on Opus 4.8, GPT-5.5, and Kimi K2.7. Every single model gave the same result. This wasn’t some unique Fable 5 flaw. Anthropic’s point was simple: why pull one product used by millions, when every rival model has the same gap?

Anti-AI sentiment is growing, and it could derail OpenAI’s IPO plans

Bad Blood with the Trump administration

This didn’t come out of nowhere. Back in February, Trump told federal agencies to drop Anthropic. The company had refused to let its tech be used for autonomous weapons or unrestricted mass surveillance. The Pentagon then called Anthropic a “supply chain risk” in March, a label it’s still fighting in court. So when the export order landed, many people saw it as round two.

The restoration

The shutdown lasted 19 days. Mythos 5 came back first, cleared on 26 June for over 100 approved US organisations through Anthropic’s Glasswing programme. Commerce lifted the export controls on both models on 30 June. Fable 5 returned worldwide on 1 July, across Claude.ai, the Claude Platform, Claude Code, and Claude Cowork. It now runs a new safety filter that blocks the reported jailbreak in over 99% of cases and sends blocked requests to Opus 4.8 instead.

Where things stand now

Fable 5 briefly went offline again on 1 July, but that was an unrelated tech glitch, fixed within hours. Here’s the real catch for regular users. Fable 5 only counts within your normal Pro, Max, Team, or select Enterprise plan up to 50% of weekly limits. And that deal ends 7 July. After that, you pay through usage credits. Anthropic says this isn’t forever. It plans to bring Fable 5 back into standard plans once it has enough capacity. For now, heavy users are watching that deadline closely.

The jailbreak isn’t the real story here. The precedent is. A government pulled a live AI model offline overnight, with no court order and no public evidence. That’s new territory. Whatever you think of Anthropic’s past safety warnings, this is the first time a “too powerful to release” claim got taken this literally, by the very people writing the rules.

---

For more info pieces like this, click here!