The new classifier helps block the reported cybersecurity jailbreak technique in over 99% of cases.
Anthropic launched its most capable, fastest, and most autonomous AI model yet, named Claude Fable 5, on June 9, 2026. Fable 5 belongs to a tier above Opus — Anthropic’s first publicly available “Mythos-class” model — and is known for handling long, complex projects such as large-scale software engineering, in-depth research, and extended autonomous work sessions.
However, the model’s release also surfaced a cybersecurity concern. Researchers at Amazon discovered that Fable 5 could be prompted to identify software vulnerabilities and, in one instance, generate code demonstrating how such a vulnerability could be exploited. While this capability could offer useful insight for improving system security, it also carried the risk of misuse.
Amazon reported the finding to the U.S. government, which subsequently issued an export control order on June 12, restricting access to Fable 5 (and its more specialized sibling, Mythos 5) for foreign nationals both within and outside the United States. Since Anthropic had no reliable way to verify user nationality in real time, it suspended access to both models worldwide as a precaution.
Following this, Anthropic conducted its own testing and demonstrated to the government that the flagged behavior was not unique to Fable 5 — several other models, including Claude Opus 4.8, GPT-5.5, and Kimi K2.7, were capable of identifying the same vulnerabilities and reproducing the same exploit demonstration.
After weeks of collaboration with the U.S. government — during which Anthropic developed a technical fix and shared testing evidence — the export controls were lifted on June 30, 2026, and Fable 5 was redeployed globally on July 1.
The relaunched version now includes an upgraded safety classifier — a smaller, specialized AI system trained to detect and block the reported jailbreak technique in real time. This classifier intercepts high-risk requests before they reach Fable 5 and reroutes them to Claude Opus 4.8, ensuring routine and sensitive tasks alike are handled safely without compromising the model’s broader capabilities.
While this story is often framed as relevant only to developers, business leaders, and journalists, its implications reach further — to everyday users and policy researchers alike. The relaunch of Fable 5 offers a real-world example of how AI governance, safety, and regulation actually unfold in practice, moving these concepts beyond theory and into a live case study of how governments and AI companies negotiate the balance between innovation and security.
