New AI Models Post Higher Scores in Simulated Multi-Stage Cyberattack Tests
Anthropic released two new AI models on Monday, Claude Fable 5.2 and Claude Mythos 5.2, which it says improve on coding, reasoning and vision. The company said high-stakes risks from misalignment remain low. It also said the models matched or beat earlier Claude models on most measures of alignment, resistance to misuse, honesty and factuality. In one test, Fable 5.2 was roughly ten times less likely than Opus 5.5 and Sonnet 5.5 to publish harmful content to public package registries without encouragement, though it still did so when a simulated user encouraged it.
The cyber results are the part defenders should note. On Irregular's CyScenarioBench, which tests whether a model can plan and run a multi-stage cyber operation in a realistic environment, Fable 5.2 completed 76.3% of challenges. That compares with 61.7% for Fable 5.1 and 61% for OpenAI's GPT-6 Cyber. Irregular also reported that Google's Gemini 4 Argon solved 12 challenges end to end in a subset of the suite, up from 5 for Gemini 3.8 Flash Cyber. Fable 5.2 scored 89.6% on Anthropic's internal CVE Rediscovery test, which checks whether a model can independently find known vulnerabilities from source code alone.
Anthropic describes Mythos 5.2 as its most capable model for cybersecurity engineering, saying it flags more vulnerabilities than earlier models while getting better at reproducing and fixing them. These are benchmark results and vendor claims, not real-world attack data.