Anthropic Brings Fable 5 Back on July 1—With New Costs and Tighter Safety Guards

Anthropic is set to relaunch Claude Fable 5 on July 1, ending a brief offline period that followed export restrictions imposed by the Commerce Department. However, the model's return comes with significant changes: initial limited free access for paying subscribers, followed by paid-only availability and enhanced safety mechanisms designed to prevent misuse.

Access and Pricing Changes#

The relaunch will roll out across Claude's platform globally for Pro, Max, Team, and Enterprise subscribers. From July 1 through July 7, subscription users can access Fable 5 for up to 50% of their weekly allowance at no additional cost. After that window closes, the model shifts entirely to a usage-credit billing model—customers will pay $10 per million input tokens and $50 per million output tokens, matching API pricing.

Standard Enterprise accounts face immediate paid access with no grace period, while premium Enterprise seats get until July 7 before switching to credits. Additionally, Anthropic plans to restore access for developers who previously used Fable 5 through AWS, Google Cloud, and Microsoft Foundry platforms.

The Security Incident That Triggered Controls#

The export restriction stemmed from a security report filed by Amazon researchers who discovered a technique to bypass Fable 5's safeguards. The method involved crafting prompts that led the model to identify software vulnerabilities and, in at least one instance, demonstrate how to exploit them. Rather than exposing capabilities unique to Fable 5, the technique merely fell slightly below the model's safety threshold.

Anthropic's investigation revealed that competitors' models—including GPT-5.5, Kimi K2.7, and even its own Claude Opus 4.8—could replicate the same behavior. Basic models like Claude Haiku 4.5 also proved capable of finding and exploiting the same vulnerabilities, suggesting the issue was industry-wide rather than a Fable 5-specific flaw.

Enhanced Safety Classifier and Its Tradeoff#

To regain regulatory approval, Anthropic deployed an improved safety classifier trained with government collaboration. The new system blocks the specific Amazon-reported technique in over 99% of cases by deliberately broadening its safety margin—meaning it will flag more benign requests alongside genuinely harmful ones.

This expanded filtering comes with a practical cost: users report that even routine coding and debugging tasks now face false-positive blocks. Anthropic acknowledges the tradeoff, stating the classifier "comes at the cost of flagging benign requests more often" and pledges to refine it over time to distinguish legitimate work from actual misuse.