[leptons]

But MRI machines can't be turned on and shut off that easily. As someone here explained, it takes up to 15 minutes for the magnet in an MRI to "shut down", and costs $50,000 each time.

Why not just control access to the room behind a metal detector? It would be really simple, but effective. I don't think any MRI should be allowed to operate without this basic level of protection.

[scarier]

Sure, an engineering control for MRI room access would be implemented differently--that's just the canonical example that people are familiar with. One possible implementation for MRI access is the airlock method, where the inner access door would only be allowed to unlock with the outer door locked and no metal detected in the space between (also the outer door would be prohibited from unlocking when the inner door is unlocked, except for some kind of inner emergency override that might also be tied to the emergency quench).

Literally no one disagrees with you on this, and most (if not all) hospital administrators will say they already do it the way you suggest. I'm pointing out that the actual implementations I'm aware of are often ineffective because they use administrative rather than engineering controls, and this is a critical distinction people need to be more aware of when interacting with dangerous systems. Managers, at least in my experience, tend to wildly overestimate compliance rates with administrative controls, even ignoring any possibility of deliberate noncompliance.