The NHS keeps adding AI but when it goes wrong, clinicians seem to be getting the blame.

What's happened

The commission spent a year on it and heard from more than 12,000 patients, staff and developers. None of it is law yet. The government will respond separately.

The bit that matters for you

Right now, if an AI tool gets something wrong, the claim lands on the clinician and the trust, because you're the ones with the clearest duty of care. The report has a name for this: a liability sink. The blame slides down to the person who used the tool, and the tool itself doesn't get looked at.

And staff told the commission the same thing. You're expected to spot when a tool drifts and catch its mistakes, without being given the information you'd need to do that. That's the nurse acting on a triage score and the pharmacist checking a prescription, as much as the doctor signing it off.

So the report asks for 3 things. Who's responsible at each stage, written down. Contracts between the manufacturer and the trust that say who does what. And guidance for you on where your responsibility starts, where it ends, and when it's OK to rely on the tool. Until then, the GMC's AI guidance is what doctors have, and the other regulators are working on theirs.

What else lands on you

1. 1You'll be expected to be competent in it. Same as any other technology in your job. Training from the start of your career through to CPD, shaped to each profession, and your employer has to train you on the specific tools you use.
2. 2Reporting a faulty tool is part of your duty of care. For doctors it already is, under Good Medical Practice. The report wants a report button built into the tool, the Yellow Card scheme made to work for AI, and the tool and its version written into the patient's record, so you can see which AI touched which patient.
3. 3New AI gets L-plates. New tools would go in under supervision first and earn full approval by performing in the real world. They'd be monitored for the whole of their working life, not approved once and forgotten. So the tool on your ward might still be on its L-plates, and what you report about it counts.
4. 4Patients will expect to be told. Accuracy is the public's top priority. They also want a human overseeing it, and to know when it's been used. The report doesn't want that left to you in every consultation. It wants a system-wide notice, a proper conversation kept for the high-risk uses, and the option to opt out where that's possible.

What staff said

77% of the healthcare professionals who responded want the rules completely overhauled or significantly reformed. And only 48% of staff think AI is safe for patient care, against 86% for admin. So people are fine with it doing the paperwork. They're not yet fine with it making the clinical call.

What's already here

AI is already in the NHS, spotting strokes and skin cancers early and taking admin off clinicians. I've seen it from the inside.

The tools work. What's been missing is the structure around them, and that's what this report is about.

Where it sits now

The government wants the NHS to be the most AI-enabled health system in the world. The report's point is that it won't get there while the people using the tools carry the blame alone. It's now with DHSC and the MHRA, and their response will say which recommendations become policy.