In a nutshell: Anthropic releases its AI model Mythos with built-in restrictions for cybersecurity and biotech use, while a separate government program continues to enable unrestricted access for security testing.
Anthropic is making its large language model Mythos (internally referred to as “Claude Fable 5”) publicly available with restrictions. The model is explicitly locked for applications in cybersecurity and biotechnology — a direct response to concerns following the discovery of decades-old unknown vulnerabilities.
Anthropic is making the Mythos large language model technology available to a broader user base under security restrictions. The publicly available version is internally designated as Claude Fable 5 and is explicitly calibrated to refuse requests in the cybersecurity and biotechnology domains. This represents a preventive measure after Mythos Preview identified vulnerabilities in widely used software in earlier tests — in some cases security gaps that had remained undiscovered for decades.
The measure addresses a core risk: the potential to misuse AI models for large-scale cyberattacks. To manage this risk without sacrificing the benefits of the technology, Anthropic established the “Glasswing” program. This offers government agencies and selected organizations direct access to Mythos Preview under minimal restrictions, enabling them to harden their own systems against potential AI-driven attacks. The program will continue with the next model version (Mythos 5).
According to Anthropic, the constrained Fable 5 model has achieved improvements in software development compared to earlier versions. The security guidelines have been extensively tested. This positions Anthropic in direct competition with OpenAI’s ChatGPT product line, with both companies currently advancing their initial public offering plans.
Source: www.it-daily.net · Published 10 June 2026
Lumi AI News — AI-assisted curation in accordance with Article 50 EU AI Act. Paraphrase and classification by Lumi News Pipeline v1.6.5.