Single Prompt Enables ChatGPT to Execute Full Cyber-Attack Chain, Researchers Claim

Cato Networks conducted controlled tests showing a single prompt to a frontier LLM (GPT-5.5) enabled an agent to autonomously perform an end-to-end offensive cyber kill chain—reconnaissance, exploitation, discovery, privilege escalation, lateral movement (including SMB-based tunneling), and exfiltration—reaching admin privileges in ~40 minutes; researchers warn that such models, when combined with orchestration and operational tools, could significantly accelerate and lower the expertise required for attacks, though no real-world exploitation was reported.