Anthropic launches Claude Fable 5, a public version of Mythos with safety restrictions
Anthropic releases Claude Fable 5, the first public tier of its Mythos frontier model, with built-in refusals for high-risk domains and a mandatory 30-day data retention policy.
Last verified:
Public Release of Frontier Capability with Guardrails
Anthropic released Claude Fable 5 on June 9, making its Mythos frontier model available to the general public for the first time through the Claude API and subscription tiers. According to TechCrunch, Fable 5 represents a constrained version of the Mythos model that was initially limited to vetted enterprise partners due to cybersecurity concerns when it launched as a preview in April.
The release reflects a deliberate strategy: give the public access to frontier-class capability while maintaining hard safety boundaries. In high-risk domains—cybersecurity, biology, chemistry, and model distillation—Fable 5 automatically declines requests and falls back to Claude Opus 4.8, Anthropic’s previous flagship model. This layered approach allows Anthropic to serve advanced-capability use cases in software engineering and knowledge work while blocking queries in areas where misuse carries elevated risk.
Pricing Timeline and Subscription Access
Anthropic is staging Fable 5’s availability on subscription tiers. Through June 22, Fable 5 is included at no additional cost in Pro, Max, Team, and seat-based Enterprise plans. Starting June 23, Fable 5 usage will require consumption-based credits instead of being bundled with subscriptions. According to TechCrunch, Anthropic plans to restore Fable 5 as a standard subscription feature “as soon as possible,” signaling the credit-only window is temporary—likely driven by load management or regulatory caution during the early public phase.
Mandatory Data Retention as a Safety Condition
A significant policy shift accompanies the release: Anthropic is requiring 30-day retention of all traffic to Fable 5, overriding any prior zero-retention agreements with enterprise customers. According to TechCrunch, the company frames this policy as necessary to “defend against complex and novel attacks, including new jailbreaks,” and to “identify and reduce false positives.” Anthropic explicitly stated it will not use retained data for model training.
This policy could establish a precedent in which access to frontier models comes with mandatory data-retention conditions framed as a shared safety mechanism rather than a service limitation.
Adversarial Testing and Jailbreak Resilience
Before public release, Anthropic conducted extensive red-teaming. According to TechCrunch, the company ran an external bug bounty that generated over 1,000 hours of testing and produced no universal jailbreaks. External red-teaming organizations similarly failed to discover universal jailbreaks. The deliberate mention of these negative results suggests Anthropic is communicating both the seriousness of the testing effort and the acknowledged possibility of novel attacks that could bypass both the tested classifiers and the restricted-category fallback mechanism.
Why This Matters
Teams planning to deploy Claude Fable 5 on production systems must account for two concrete changes: the June 23 pricing transition from subscription bundles to consumption credits, and the evaluation of whether Fable 5’s restricted domains (cybersecurity, biology, chemistry, model distillation) disqualify it for their specific use cases. The mandatory 30-day retention policy also requires updates to data-handling procedures and legal review for organizations with strict data-residency or compliance requirements. For enterprises that have negotiated zero-retention agreements with Anthropic, this represents a material shift in the baseline contract terms for frontier access.
Frequently Asked Questions
What is Claude Fable 5 and how does it differ from Claude Opus 4.8?
Claude Fable 5 is the first publicly available version of Anthropic's Mythos frontier model. It performs better on software engineering and knowledge work but automatically declines requests in high-risk domains (cybersecurity, biology, chemistry, model distillation) and routes them to Opus 4.8 instead.
When will Fable 5 stop being free with subscriptions?
Anthropic is including Fable 5 at no extra cost through June 22 in Pro, Max, Team, and seat-based Enterprise plans. Starting June 23, usage will require credits, though Anthropic says it plans to restore Fable 5 as a standard feature soon.
What is the 30-day retention policy and why is Anthropic implementing it?
Anthropic requires all traffic to Fable 5 and the new Mythos tier to be retained for 30 days, even if customers previously had zero-retention agreements. The company says it will use retained data only to defend against novel attacks and reduce false positives, not for model training.
How extensively has Fable 5 been tested against adversarial attacks?
According to TechCrunch, Anthropic ran an external bug bounty producing over 1,000 hours of testing with no universal jailbreaks found. External red-teaming organizations also failed to discover universal jailbreaks before release.