Anthropic’s Flagship Returns This July – Here’s What You Get (and What Changed)

Most people assumed that once a government switches off an AI model, it stays off for a long time. Fable 5 just proved otherwise: nineteen days after the US ordered it dark, Anthropic’s flagship is live again for everyone.

What actually happened

On June 30, 2026, the Commerce Department lifted the export controls it had slapped on Claude Fable 5 and its sibling Mythos 5 back on June 12. One day later, on July 1, Anthropic flipped the switch: Fable 5 is once again available globally on Claude.ai, the Claude Platform, Claude Code, and Claude Cowork, with AWS, Google Cloud, and Microsoft Foundry access being re-enabled as quickly as possible.

The original shutdown was never about a broken product. The June order required Anthropic to block foreign nationals everywhere, including its own non-citizen staff, and with no way to verify nationality in real time, the company pulled the model for everyone instead.

What triggered the whole mess

A jailbreak. Amazon researchers found a prompt that got Fable 5 to identify software vulnerabilities and, in one case, produce code showing how one could be exploited. That report reached the government, and the export controls followed.

Here is the twist: Anthropic’s own testing found that plenty of less capable models, including Opus 4.8, GPT-5.5, and Kimi K2.7, could identify the same vulnerabilities, and every model tested could reproduce the exploit demonstration. In other words, nothing in the report showed a uniquely dangerous capability. The government eventually agreed, and the restrictions came off.

The version coming back is not identical

Anthropic did not just redeploy the old model. Working with the government, it trained an improved safety classifier that targets and blocks the behaviour described in the report. If a request trips the classifier, you will see a notification and the request gets handed to Claude Opus 4.8 instead.

The catch: in the near term, some routine work, including ordinary coding and debugging, may trigger that fallback. Early reports from developers in the first days back describe a mixed experience in Claude Code, with occasional Opus 4.8 handoffs on tasks that seem harmless. Anthropic says it is tuning this, and it has opened a HackerOne bounty program for researchers who find new jailbreaks.

How access and pricing work in July

There is a limited-time sweetener. Through July 7, Fable 5 is included in Pro, Max, Team, and select Enterprise plans for up to 50% of your weekly usage limits , no extra cost. After July 7, it moves to usage credits.

A few details worth knowing:

• API pricing is unchanged: $10 per million input tokens and $50 per million output tokens.
• Standard Enterprise seats get no included allowance, usage credits must be enabled or users simply will not see the model.
• The mandatory 30-day data retention rule still applies, and zero-data-retention is not available.
• Mythos 5, the unrestricted sibling, remains limited to vetted organisations only.

The Simple Takeaway

Fable 5’s July return closes out the strangest three weeks in frontier AI so far: a launch, a government shutdown, and a redeployment, all in under a month. The model is back, the price did not move, and the first week is effectively a free trial inside existing paid plans. If you want to test the most capable generally available model on the market, the window before July 7 is the cheapest way to do it.

Frequently Asked Questions

When exactly did Fable 5 come back? July 1, 2026, one day after the Commerce Department withdrew its export controls.

Where can I use it? Claude.ai, the Claude Platform (API), Claude Code, and Claude Cowork, with AWS Bedrock, Google Cloud Vertex AI, and Microsoft Foundry following shortly.

Is it free to try? If you are on Pro, Max, Team, or select Enterprise plans, it is included for up to 50% of your weekly usage limits until July 7. After that, it runs on usage credits.

Did the model get weaker? The underlying model is the same. What changed is a stricter safety classifier that can block certain requests and route them to Opus 4.8, occasionally catching routine coding tasks while Anthropic tunes it.

What about Mythos 5? Still restricted to approved organisations. Everyone else gets the same capabilities through Fable 5, just with the safety layer attached.