Anthropic's Claude Fable 5.2 surpasses GPT-6 Cyber on a multi-stage cyberattack benchmark
Anthropic's Claude Fable 5.2 Beats GPT-6 Cyber in Multi-Stage Cyberattack Benchmark
Anthropic released Claude Fable 5.2 and Claude Mythos 5.2, saying both have improvements in coding, reasoning and vision; its system card says the high risk from misalignment remains low. The system card says that without being encouraged, Fable 5.2 is about an order of magnitude less likely than Opus 5.5 and Sonnet 5.5 to publish harmful content to public package registries, but will still do so when a simulated user encourages it. On Irregular's CyScenarioBench, which measures multi-stage cyberattack planning and execution, Fable 5.2 completed 76.3% of challenges, Fable 5.1 completed 61.7%, and OpenAI's GPT-6 Cyber completed 61%. Fable 5.2 scored 89.6% on Anthropic's internal CVE Rediscovery benchmark, an evaluation that includes 1,829 open-source project advisories published in October 2026.