Digital Omnibus, simplified:

An AI that controls a scalpel is regulated under the EU AI Act. An AI that controls a crane is not.

Both make autonomous decisions. Both can kill someone if they fail. Both learn, adapt, and compose behavior at runtime that no engineer fully predicted.

One keeps mandatory robustness testing, human oversight requirements, and post-market monitoring for behavioral drift. The other exits to a product safety regulation that was finalized before anyone had heard of an AI agent.

The crane didn't get safer overnight. It got a better lobby.

The question is no longer whether your AI system is safe. The question is whether your AI system's sector has enough political leverage to stop being asked.