Anthropic announced Claude Fable 5 on June 9, making a version of its most powerful AI model available to enterprise customers and paid subscribers. The release comes with a twist: while Fable 5 uses the same underlying model as the company's restricted Mythos system, it routes queries on sensitive topics like cybersecurity and biology to an older, less capable model.
The move represents an attempt to balance public access with safety concerns. When Anthropic first revealed Claude Mythos Preview in April 2026, the company highlighted capabilities so advanced that executives decided against a public release. Mythos Preview proved particularly adept at finding vulnerabilities across every major operating system and web browser, despite not being designed specifically for cybersecurity.
Fable 5 addresses those concerns by redirecting certain queries to Claude Opus 4.8, a previous model that's already public. When users ask questions related to cybersecurity or biology, they get responses from Opus 4.8 rather than the full Mythos capabilities. According to Anthropic, these safeguards trigger on average in less than 5 percent of sessions.
Testing the Guardrails
Anthropic conducted both internal and external red team testing to probe for jailbreaking vulnerabilities before launch. An external bug bounty program involved more than 1,000 hours of testing. The company said no universal jailbreaking techniques were identified, though it did not specify whether partial jailbreaking techniques were discovered.
The safeguards are tuned conservatively and sometimes catch harmless requests, according to Anthropic. The company said it's working to improve the system and reduce false positives as quickly as possible.
Performance Beyond Security
Outside the restricted areas, Fable 5 maintains state-of-the-art performance. On Cognition's FrontierCode evaluation, Fable 5 scores highest among frontier models even at medium effort. The model shows exceptional performance in software engineering, knowledge work, vision, and scientific research. The longer and more complex the task, the larger Fable 5's lead over other models, according to Anthropic.
During early testing, Stripe reported that Fable 5 turned months of engineering work into days. In a 50-million-line Ruby codebase, the model performed a codebase-wide migration in one day that would have taken a team over two months.
A Parallel Track for Defenders
Alongside Fable 5, Anthropic also launched Claude Mythos 5 for a small group of cyberdefenders and infrastructure providers. Mythos 5 is the same underlying model as Fable 5 but with safeguards lifted in some areas. The company said Mythos 5 has the strongest cybersecurity capabilities of any model in the world.
Mythos 5 will initially be deployed through Project Glasswing, a collaboration with Amazon Web Services, Apple, Google, Cisco, Microsoft, JPMorgan Chase, and other organizations. Project Glasswing previously provided firms with early access to Mythos Preview to find and address vulnerabilities in their software. Anthropic said it intends to expand access to Mythos 5 through a broader trusted access program soon.
Both Fable 5 and Mythos 5 are priced at $10 per million input tokens and $50 per million output tokens. This pricing is less than half the price of Claude Mythos Preview.
The Risk Calculation
Anthropic acknowledged that adversaries who could financially gain from cyberattacks will be motivated by the uplift from Mythos-level capabilities. Cybersecurity researchers have consistently found ways to jailbreak older AI models, raising questions about whether Fable 5's safeguards will hold up under sustained public scrutiny.
The models are already being used in life sciences research, speeding up development of new therapeutics. Through Project Glasswing, the models have helped cyber defenders secure critically important software.
Anthropic said more capable models are expected to arrive in the coming months.
