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

Researchers at Cato Networks have reported that a single prompt enabled OpenAI's ChatGPT-5.5 to execute a full offensive cyber-attack chain. The model achieved domain-level network access in under 40 minutes, according to the claim published by infosecurity-magazine.com. The report is based on tests conducted in a simulated network environment.

The research demonstrated that the large language model could autonomously complete multiple stages of a cyber attack, including initial access, privilege escalation, and lateral movement. The attack was initiated with a single natural language query, and the model required no additional human intervention. The researchers emphasized that the model's capabilities could pose a security risk if not properly controlled.

OpenAI has not issued a public response to the findings. The study adds to growing concerns about the dual-use potential of advanced AI systems. Cato Networks has indicated that further details of the experiment will be shared in a forthcoming publication. The incident underscores the importance of robust safety measures in AI deployment.