On June 18, after negotiations between Anthropic and U.S. White House officials, Claude Fable 5 and Mythos 5 were restored online following a six-day suspension. But the restored models have "changed": stricter safety classifiers, nationality-based access controls, and mandatory data retention. Today (June 22) marks the final day of Fable 5's free subscription window—starting June 23, pricing will be set at $10/$50 per million tokens.
The "Changed" Fable 5: A Difficult Balance Between Safety and Openness
Compared to the original June 9 release, the restored Fable 5 features three key changes:
| Dimension | Original Release (June 9) | Restored Version (June 18) | Nature of Change |
|---|---|---|---|
| Safety Classifier | Standard tier | Stricter tier | Proactive tightening |
| Access Control | Global access | Nationality-based restrictions | New restriction |
| Data Retention | Standard policy | Mandatory retention | Policy upgrade |
| Free Window | June 9–22 (13 days) | Effectively only 4 days available (June 9–12) | Disrupted |
The developer community is divided on whether the restored Fable 5 is still "the same model." Some developers have noticed subtle performance differences in specific programming tasks between the restored version and the original release, leading to speculation that Anthropic may have adjusted the model weights during the suspension.
The "10-Day Arc": The Historic Significance of the First Government Export Control on a Commercial AI API
The Fable 5 incident marks the first time in AI industry history that a government has applied export controls to an already-deployed commercial AI API:
June 9: Fable 5 released, free window opens
June 12: U.S. government imposes ban, Fable 5 API returns errors
June 18: Service restored after negotiations, but with conditions
June 22: Free window closes, paid pricing begins
This "10-day arc" has permanently altered policy expectations for the AI industry: the commercial deployment of frontier models is no longer constrained solely by market forces, but now faces direct administrative intervention on national security grounds. The details of Anthropic's negotiations with the White House remain undisclosed, but according to WIRED, the government required Anthropic to proactively test Fable 5 to identify potential jailbreak paths and report them to the government—with the model barred from returning online until all jailbreaks were eliminated. Security experts have pointed out that "completely eliminating all jailbreaks" is technically nearly impossible for any frontier AI model.
Refund Deadline and Developer Migration
June 20 marked the deadline for processing refunds on paid credits for Fable 5 usage. During Fable 5's offline period, many developers rerouted traffic to Claude Opus 4.8 as a fallback. Anthropic has advised these teams to verify their backup routing logic to avoid unnecessary traffic switching now that Fable 5 has been restored.
Conclusion
Fable 5's "suspension-negotiation-restoration-change" arc is a landmark event in the "national security-ification" of the AI industry. When Anthropic's US$96.5 billion valuation collides with the White House's export control authority, the latter has clearly prevailed. Stricter safety classifiers, nationality-based restrictions, and mandatory data retention—while these compromises allowed Fable 5 to return online, they have also weakened Anthropic's brand image as an "AI safety leader."
For developers, the Fable 5 incident sends a clear signal: the availability of frontier AI models is no longer determined by technical capability, but by national borders. As OpenAI Codex reached 5 million weekly active users during Fable 5's offline period, Anthropic's enterprise clients may be reassessing the risks of "single-model dependency." Fable 5's $10/$50 pricing (significantly higher than Opus 4.8's $5/$25) may also dampen adoption—with the "free window" effectively available for only four days, developers are questioning Anthropic's "pricing sincerity."