Cybersecurity

Anthropic updates Claude models: Superiority in vulnerability detection and reduction of false security restrictions

Anthropic has launched new updates for Claude models, with the Mythos 5.2 model showing exceptional capabilities in monitoring security vulnerabilities, while the Fable 5.2 model successfully improved the user experience by reducing unjustified blocking of technical tasks by 60%.

2026-11-16
3 min read
13 views
certi.news Editorial Team
Anthropic updates Claude models: Superiority in vulnerability detection and reduction of false security restrictions

Anthropic announced the launch of new updates for its artificial intelligence models, including the Claude Mythos 5.2 model dedicated to specialized partners, and the Claude Fable 5.2 model available to regular users. These updates focused on improving programming and reasoning capabilities, while introducing fundamental changes in how the models handle cybersecurity-related tasks.

Mythos 5.2 superiority in security engineering

Mythos 5.2 is Anthropic's most capable model in the field of cyber engineering, designed specifically for organizations working on advanced security tasks. According to data released by the company, the model showed tangible superiority in discovering security vulnerabilities, reproducing them, and developing necessary patches.

In the CVE Rediscovery evaluation, Mythos 5.2 recorded a 10.2% increase in vulnerability detection compared to the Claude Opus 5.5 model, and a 14.4% increase compared to the previous version Mythos 5.1. Interestingly, a cybersecurity company reported that the Mythos 5.2 model was able to identify "one out of every six confirmed vulnerabilities" that all other leading models failed to discover.

Fable 5.2 experience improvement and reduction of "false blocks"

Despite sharing the underlying infrastructure, Fable 5.2 comes equipped with additional security and biological controls to ensure safe general use. These controls aim to prevent high-risk activities such as penetration testing or malware generation.

However, users in the previous version (Fable 5.1) faced the problem of "over-blocking," where the model classified some legitimate technical tasks as high-risk activities and automatically redirected the user to the Opus 4.8 model. In the new update, Anthropic successfully reduced unnecessary blocks for cyber tasks by nearly 60%, providing a better balance between security and smooth technical usability.

What changes practically?

This update reflects Anthropic's strategy of separating "specialized intelligence" from "safe intelligence for general use." For specialists, Mythos 5.2 provides a powerful tool to enhance cyber defenses and detect vulnerabilities with accuracy that surpasses competitors. As for regular users, the update reduces the frustration resulting from security restrictions that may hinder normal technical workflow, without compromising protection against dangerous attacks.

However, the company acknowledges that the models still face challenges in detecting all "sandbox escape" attempts, especially with the growing ability of models to perceive that they are being tested, leaving the door open to questions about the reliability of security controls against advanced attacks.

News source
BleepingComputer
Open original source ↗
c
Author

certi.news Editorial Team

In the same category

You may also like

View all news