Some of the most advanced artificial intelligence models were able to develop cyberattack plans targeting critical infrastructure, including water and energy supply systems, in approximately 90% of attempts. The finding comes from a study by cybersecurity company Frenos published on Wednesday.
Frenos tested 10 AI models for their ability to identify vulnerabilities, assess access requirements and plan attacks on industrial systems. In 16,980 tests, three leading commercial models produced workable attack plans in 89.5–92.9% of attempts. The models’ names were not disclosed.

However, the researchers stressed that the tests were conducted using supplied data and simulated scenarios rather than real-world conditions. The models sometimes skipped important steps or assumed access they did not have.
The company urged critical infrastructure operators to strengthen network segmentation, restrict remote access and monitor suspicious activity.