A UK expert statement says failing to use AI can itself be negligent, and the text we can verify is the January draft

What we opened, and what we did not

This piece reports the UK Jurisdiction Taskforce's draft Legal Statement on liability for non-deliberate AI harms, published for consultation in January 2026 and introduced by Sir Geoffrey Vos, Master of the Rolls. That document is public and we read it.

A final version was published on 7 July 2026. It is distributed only behind a registration form, and the UKJT resource page states that no PDFs have been added to that resource. We have not opened it, so we do not quote it and we do not report what it holds.

Every quotation below is from the January 2026 draft. Where the final differs, and there is reason to think it does in at least two places, this piece does not cover it. That limitation is set out again at the end.

What the statement is, and who wrote it

The UKJT describes the work as "an authoritative Legal Statement on liability for non-deliberate AI harms under English private law." It is an expert statement of existing law, not an instrument. Nothing in it binds a court.

The drafting team was Matthew Lavy KC, Richard Munden, Lucy McCormick, Iain Munro, Isabel Barter and Jacob Turner, working with a wider expert group that included Prof Ryan Abbott, Lawrence Akka KC, Matt Frank, Prof Sarah Green, David Quest KC, Tom Whittaker, Prof Peter Wills and Michael Workman.

UKJT statements carry weight because of who prepares them, not because they have legal force. We are not aware of an English court having cited this one.

The draft's starting premise: AI is never the defendant

The draft sets out its foundation plainly. Its approach "starts from the premise that AI does not have legal personality in English law and that, therefore, it cannot itself be held legally responsible for physical or economic harm; instead, liability for harms that arise from the use of AI must be attributed to legal persons, using ordinary legal principles."

That is the drafting team's stated premise, and it drives their analysis of vicarious liability, the doctrine that makes an employer answerable for an employee's wrongs. Because vicarious liability "involves Anita being liable for the torts of another person," the draft reasons, "it is not capable of causing her to be liable for the actions or failures of an AI because AI does not have legal personhood."

The draft leaves a human route open. It accepts that a person could still be vicariously liable where the harm arose because a human employee acted negligently and thereby caused the AI harm.

The line that should change a partnership meeting

On professional negligence, the draft states that if a professional acts negligently in relation to AI use, "whether by using AI where it should not have been used; or using inappropriate AI; using AI with insufficient care; or indeed failing to use AI when it should have been used," and that negligence causes an affected third party loss, "the professional can expect to be held liable."

The third clause is the one to sit with. Failing to use AI when it should have been used.

Almost every AI governance policy written in the last two years rests on one assumption: that the risk lives in adoption, and that caution is the safe posture. The draft puts a second risk on the other side of the ledger. If a competent member of your profession would have used a tool and you did not, the omission is available to a claimant as a breach of the standard of care.

That is an argued position in a non-binding draft, not a holding. No English court has decided it as far as we know, and we make no prediction about how one would.

What this means for a US professional

England and Wales is not most US readers' jurisdiction, so the direct effect is nil. The indirect value is worth an hour.

If your firm has England and Wales exposure, through a UK office, UK-governed engagement letters or UK clients, this draft is the most developed treatment of the question we have been able to verify. The July final supersedes it.

More broadly, it is a worked example of how a mature common-law system resolves AI liability using contract and negligence alone, without an AI statute. The US is approaching the same problem with the same tools, so the reasoning travels even though the authority does not.

For anyone underwriting or buying professional indemnity cover, the failure-to-adopt theory is worth naming in a renewal conversation. We have no evidence that any insurer has changed underwriting, or that any claim has been brought on this basis. What exists is an argument on the record.

What we could not verify

Commentary on the final July 2026 statement reports that it goes further than the draft in at least two places, on non-delegable duties and on material contribution to causation.

We are not reporting either, because we have not read the final. We can say one thing precisely: the phrase non-delegable does not appear anywhere in the January 2026 draft we opened. That is consistent with those points being new, and consistent is not verified.

If you need those holdings, request the final PDF from the UKJT and read it yourself. Do not take them from this piece, and treat any summary of them with the same caution.

Related briefings

• UK judiciary guidance on AI for judicial office holders
• The EU Product Liability Directive and AI software
• AI Regulation News hub