
Anthropic updates Claude models with enhanced cybersecurity capabilities - SC Media
Anthropic has released two new versions of its Claude AI models, Claude Mythos 5.2 and Fable 5.2, with significant enhancements for cybersecurity.
The new models, announced on November 16, 2026, build upon the company's existing Claude family with targeted improvements for security professionals and general users.
Claude Mythos 5.2 is designed as the company's most capable model for cybersecurity engineering.
It has been shown to detect 10.2% more vulnerabilities than the Claude Opus 5.5 model and 14.4% more than its own predecessor, Mythos 5.1.
Security partners reported that one in six bugs identified by Mythos 5.2 were missed by all other tested frontier models.
The second model, Fable 5.2, focuses on safety and usability.
It includes strict safeguards to prevent high-risk activities like exploit generation and unauthorized penetration testing.
When the system detects a potentially dangerous request, it automatically redirects the task to the Opus 4.8 model.
To improve the user experience, Anthropic has reduced unnecessary blocks by 60% compared to the previous Fable 5.1 version.
Balancing safety and utility
The company acknowledges that as AI capabilities grow, detecting potential failures during pre-release evaluations becomes increasingly complex.
These updates reflect the broader industry trend toward creating specialized AI models that can balance the high-level utility required for security research with robust safety guardrails for general users.