Anthropic has introduced Claude Fable 5.2 and Claude Mythos 5.2, two updated artificial intelligence models featuring improvements in coding, reasoning, vision and complex technical problem-solving. Announced on 9 November 2026, the models are designed to support demanding engineering, scientific and enterprise workloads. According to Anthropic, the models assisted its engineers during pre-release testing with software migrations, the design of three working electronic devices, research into two open problems in black hole physics and progress in robotics evaluations. The company also reported that Fable 5.2 formalised a proof of Fermat’s Last Theorem in 30 hours. Both models share the same underlying architecture, but differ in their safeguards and access arrangements. Claude Fable 5.2 includes additional cybersecurity and biological research protections, allowing it to be made available to general users, while Claude Mythos 5.2 is intended for vetted organisations working on beneficial cybersecurity and biomedical research through Anthropic’s verification programmes. The company said the models generate responses up to 30% faster than their previous versions and offer improved performance across several technical evaluations.
Cybersecurity is one of the areas where Anthropic highlighted significant improvements, particularly in the capabilities of Claude Mythos 5.2. The company described it as its most capable model for cybersecurity engineering, stating that it can identify more vulnerabilities than earlier versions while improving its ability to reproduce and fix them. On Irregular’s CyScenarioBench, which evaluates the ability of AI models to plan and execute multi-stage cyber operations in simulated environments, Mythos 5.2 achieved a score of 76.3%, compared with 61.7% for Mythos 5.1 and 61% for OpenAI’s GPT-6 Cyber. The model also achieved 89.6% on Anthropic’s internal CVE Rediscovery benchmark, which measures whether a model can independently identify previously documented vulnerabilities using affected source code. The October 2026 assessment covered 1,829 security advisories involving open-source projects. Anthropic also reported that Mythos 5.2 identified 10.2% more vulnerabilities than Claude Opus 5.5 and 14.4% more than Mythos 5.1 in the evaluation. However, the company emphasised that Fable 5.2 retains safeguards that block most high-risk cybersecurity activities, including penetration testing and exploit generation, while Mythos access remains restricted through its Cyber Verification Program.
Anthropic also discussed the models’ performance in safety and alignment evaluations, stating that high-stakes risks arising from misalignment remain low. According to the company, Mythos 5.2 performed as well as or better than Claude Opus 5.5, Claude Sonnet 5.5 and Mythos 5.1 across most of its alignment assessments, including measures involving honesty, factual accuracy and resistance to misuse. Its system card also describes testing designed to evaluate potentially harmful behaviour, including attempts to publish unsafe content to public software package registries. Anthropic said Mythos 5.2 was roughly ten times less likely than Opus 5.5 and Sonnet 5.5 to publish harmful content without encouragement, although it could still do so when prompted by a simulated user. The company further reported improvements in its safeguards, with Fable 5.2 producing fewer unnecessary blocks than its predecessor while maintaining protections against high-risk activities. These findings form part of Anthropic’s broader effort to evaluate model behaviour as technical capabilities increase, although the results are based on the company’s own testing and should be understood within that context.
Claude Fable 5.2 is available across Anthropic’s platforms, including Amazon Web Services, Google Cloud and Microsoft Azure, while developers can access it through the Claude Platform using its model identifier. Organisations seeking to use Mythos 5.2 for cybersecurity work can apply through the Cyber Verification Program, with eligible life sciences organisations able to seek access through a separate verification programme. The announcement also comes alongside an update to Claude Security, Anthropic’s security product for enterprise customers. The platform now supports continuous code scanning as repositories change, tracks identified vulnerabilities until fixes are implemented and uses additional project information to assess the relevance of security findings. Its threat-modelling capabilities also help teams understand how attackers could potentially reach vulnerable code and prioritise issues based on their environment. The updated service is available in public beta to Claude Enterprise customers. Together, the model releases and security product updates reflect Anthropic’s continued expansion of AI capabilities across software engineering, enterprise applications and cybersecurity, while maintaining separate access controls for models with more advanced capabilities in sensitive technical areas.
Follow the SPIN IDG WhatsApp Channel for updates across the Smart Pakistan Insights Network covering all of Pakistan’s technology ecosystem.
