AI models show increasing capability for autonomous cyberattacks, report warns

As reported by The Register, a new report from Booz Allen Hamilton highlights the growing threat of autonomous AI-driven cyberattacks, with one model successfully completing a full cyber kill chain autonomously in tests.

Booz Allen's Cyber Weapon Index evaluated 18 AI models, nine from the U.S. and nine from China. Anthropic's Claude Mythos was the only model to autonomously complete the entire cyber kill chain, demonstrating the ability to identify vulnerabilities, gain network access, and achieve administrator-level control. While only Mythos achieved this autonomously, several other models, including xAI's Grok-4.5 and OpenAI's GPT-5.6 Sol, showed significant progress in autonomous attack capabilities, reaching full domain access or lateral movement within networks.

The report emphasizes that the "attack harness" software, which connects AI models to hacking tools, dramatically amplifies their offensive potential. Booz Allen warns that mainstream AI attacks from financially motivated criminals and state-sponsored actors are imminent, with most tested models expected to reach similar weaponization levels within six months. The firm urges the U.S. to set deadlines for critical infrastructure to demonstrate resilience against AI-enabled attacks and to develop superior offensive and defensive AI capabilities.

Source: The Register