The bottom line: Anthropic publicly releases the more powerful Claude variant Fable 5, but automatically routes potentially dangerous cybersecurity requests to a weaker model.
Anthropic has introduced two new models based on its Mythos architecture: Claude Fable 5 is made publicly available, while Claude Mythos 5 remains restricted to around 150 selected cybersecurity and infrastructure partners. Fable 5 is intended to be the most capable publicly accessible Claude version to date, but includes safeguards against misuse for offensive cyber operations.
Anthropic describes Fable 5 as its most capable publicly available variant to date. The model outperforms earlier Claude versions in software engineering, scientific research, image processing, and complex knowledge-work tasks. The performance advantage increases with the complexity and length of tasks, enabling users to delegate more extensive projects with less direct oversight.
The Mythos model was introduced in April with access for approximately 50 recipients, as its capabilities in vulnerability discovery and offensive cyber operations raised security concerns. Anthropic has now expanded this access to 150 organizations. To make Fable 5 more broadly available without incurring misuse risks, Anthropic uses security classifiers. They automatically route requests from defined categories — cybersecurity, biology, chemistry, and model distillation — to the weaker model Claude Opus 4.8. According to Anthropic, this occurs in fewer than 5 percent of sessions.
Early testing by security researchers, however, suggests that the cyber safeguards cast a wider net than described. Rob T. Lee, Chief AI Officer at SANS Institute, reports that routine tasks involving incident response, detection, and forensic workflows were routed to Opus 4.8 during his testing. This could indicate that the classifiers broadly identify cybersecurity requests rather than distinguishing between benign and malicious activities.
Anthropic deliberately describes the safeguards as conservatively designed. The company has decided to prioritize security over usability while it continues to refine the system. Extensive internal and external testing has uncovered no consistently effective jailbreak methods that would systematically circumvent the safeguards.
Source: www.csoonline.com · Published June 9, 2026
Lumi AI News — AI-assisted curation in accordance with Art. 50 EU AI Act. Paraphrasing and classification by Lumi News Pipeline v1.6.5.