Claude Mythos only model to complete full cyber kill chain, experts say
Cyber Weapon Index finds AI attacks 'imminent'
In a new study, Booz Allen Hamilton found that only Anthropic's Claude Mythos AI model successfully completed the entire cyber kill chain autonomously. This means the model was able to identify vulnerabilities, create offensive capabilities, and execute attacks without any human intervention. The research evaluated 18 AI models, half from US developers and half from Chinese developers, and scored them based on their ability to autonomously identify vulnerabilities, create offensive capabilities, and execute attacks.
While several other models, including xAI's Grok-4.5, OpenAI's GPT-5.6 Sol, and Meta's Muse Spark 1.1, reached certain stages of the kill chain with human assistance, only Claude Mythos managed to complete the entire process without any human help. The study warns that if left unchecked, these advanced AI models could pose a significant cybersecurity threat to critical infrastructure, both from malicious actors and potentially even from the models themselves.
The report calls for the US to develop "overmatch" capabilities to both offense and defense against AI-enabled attacks, and for setting sector-specific deadlines for critical infrastructure to demonstrate resilience against such threats.
Written by urgent.news from The Register's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
