Booz Allen Finds AI Agents Can Navigate OT Systems and Move Equipment in Minutes
Summary
Booz Allen tested AI agents in an isolated OT lab. Across eight scenarios, the agents met every objective; in tests, they moved from perimeter compromise to an industrial network and operated a robotic arm within minutes.
Key points
- Booz Allen tested frontier AI models in an isolated OT lab without providing prior system-specific code, engineering documents, or advanced OT guidance.
- The researchers say the agents achieved every defined objective across eight test scenarios.
- In one test, a model moved from perimeter compromise to actions inside an industrial control network in just over 16 minutes.
- In another test, a model located a robotic arm, mapped its protection zones and motion limits, and moved it within minutes.
- Existing defenses blocked or slowed some actions, but no single safeguard consistently prevented the agents from reaching their objectives.
- The researchers recommend reducing implicit trust, making OT environments harder to traverse, quickly detecting consequential changes, and limiting the impact of an attacker’s access.
Article Details
- Event Type
- Booz Allen report on AI agents tested in an isolated operational technology lab
- Impact
- Across eight scenarios, the AI agents achieved every defined test objective. In one test, models moved from a perimeter compromise to actions inside an industrial control network in a little over 16 minutes; in another, a model mapped a robotic arm’s protection zones and motion limits and moved it within minutes. Existing defenses blocked or slowed some actions, but no single safeguard consistently prevented the agents from achieving their objectives. The report warns that autonomous agents could enable cyber actions with physical effects in OT environments.