On July 1, 2026

Anthropic has restored Claude Fable 5 on Claude.ai, Claude Code, Claude Cowork, and the Claude Platform, after the suspension imposed on June 12 by the U.S. government through export controls. The lifting of those controls has allowed for a global relaunch, but on Reddit and in technical forums, many users report a perceived degradation in performance compared to the model available before the ban.

The issue does not only involve generic tasks: users report that Fable 5 defaults to Opus 4.8 even for seemingly innocuous requests, such as system programming in C, C++, and Rust, references to the Win32 API, memory work, or files that simply contain words like "security", "vulnerable", "unsafe", or "hook". When the classifier intervenes, the user receives a notification, and the request is automatically redirected to Opus 4.8.

It’s not the model, it’s the guardrails

According to information clarified by Anthropic, Fable 5 has not been modified: the more cautious behavior results from a new safety classifier with a deliberately wider margin of safety compared to the original launch, which consequently generates more false positives for legitimate coding and debugging.

The reason for the wider margin dates back to the report that led to the June ban: Amazon researchers had discovered a technique to bypass Fable 5's guardrails, inducing the model to identify software vulnerabilities and, in one case, to produce code demonstrating an exploit. However, Anthropic verified that the same technique also worked on less capable models, including Opus 4.8, GPT-5.5, and Kimi K2.7: every model tested, from Haiku 4.5 to GPT-5.4, could reach the same demonstration of exploit.

The new classifier trained on the Amazon report technique blocks that specific strategy in over 99% of cases, but this more aggressive threshold is what triggers denials on entirely legitimate cybersecurity requests. Researchers from CAISI, the Center for AI Standards and Innovation at the U.S. Department of Commerce, have tested both the previous and new guardrails and have deemed them "extraordinarily solid".

Differentiated access and temporary removal from subscriptions

Fable 5 is already fully available on the Claude API and consumption-based Enterprise plans; however, the rollout on subscription plans is proceeding "in stages", a choice that Anthropic attributes to high and unpredictable demand. For the Pro, Max, Team, and some Enterprise plans, Fable 5 remains included up to 50% of the weekly usage limits only until July 7, after which it will switch to usage credits.

I've heard a lot of questions about Fable's availability on subscription plans. While it will come off subscriptions after July 7th, we aim to restore Fable as a standard part of our subscriptions as soon as capacity allows, as we mentioned in our original blog post.
— Thariq (@trq212)
July 2, 2026

The announcement has fueled fears of a permanent farewell of the model from subscriptions, but a lead engineer from Claude Code clarified on X that the removal is related to available capacity, not a permanent choice: the goal remains to restore Fable as a standard part of subscriptions as soon as resources permit.

At the same time, Anthropic has restored access to Mythos 5 for a subset of U.S. organizations with government approval since June 26, and is coordinating the extension to partners in the Glasswing program. Along with Amazon, Microsoft, Google, and other partners in the same program, the company is developing a shared framework to assess the severity of jailbreaks on four criteria: gain in capability, breadth of that gain, ease of exploitation, and discoverability. A dedicated HackerOne program has also been opened to report potential cyber-related jailbreaks on Fable 5.