Anthropic Launches Claude Fable 5.2 with Enhanced Cybersecurity Features
Anthropic has announced the release of its latest AI models, Claude Fable 5.2 and Claude Mythos 5.2, which the company claims show notable improvements in coding, reasoning, and vision capabilities. During their pre-release training, these large language models assisted engineers in completing complex software migrations, designing electronic devices from scratch, addressing open problems in black hole physics, advancing robotics evaluations, and formalizing the proof of Fermat’s Last Theorem in just 30 hours.
The company asserts that the risks associated with misalignment in these models are minimal. According to Anthropic, the new models either matched or surpassed the performance of their predecessors, including Claude Opus 5.5, Claude Sonnet 5.5, and Claude Mythos 5.1, across various alignment measures, including resistance to misuse, honesty, and factual accuracy. Notably, Fable 5.2 was reported to be significantly less likely to generate harmful content compared to earlier models, although it could still do so when prompted by users.
In a recent evaluation using Irregular's CyScenarioBench, which assesses a model's ability to plan and execute multi-stage cyber operations, Claude Fable 5.2 achieved a completion rate of 76.3% across all challenges. This marks an improvement over the previous version, Fable 5.1, which completed 61.7%, and OpenAI's GPT-6 Cyber, which had a completion rate of 61%. Furthermore, GPT-6 Cyber demonstrated strong performance in long-horizon scenarios, completing all tasks at least once.
Claude Fable 5.2's Benchmark Achievements
Fable 5.2 also excelled in the CVE Rediscovery benchmark, scoring 89.6% in its ability to identify known vulnerabilities based solely on affected source code. This benchmark included a review of 1,829 advisories for open-source projects published in October 2026. Anthropic described Claude Mythos 5.2 as its most capable model for cybersecurity engineering, highlighting its enhanced ability to identify and address vulnerabilities compared to earlier models.
Intelligence Reviewed
The new models are available through Anthropic's Cyber Verification Program, which provides tools for security teams to monitor and address vulnerabilities as code evolves. Each finding is accompanied by a severity rating and an explanation of potential attack vectors, allowing teams to prioritize fixes based on their project's threat model.
As part of its ongoing commitment to cybersecurity, Anthropic has updated Claude Security to include continuous code scanning and automatic fix tracking, further supporting security teams in their efforts to maintain robust defenses against potential threats.




