AI got it wrong. Who pays? Two years on, the answer is in

Key developments, 2025 to 2026

In June 2025, lawyers put fake or inaccurate citations of cases before the court. The cases had been generated by AI and simply did not exist.

In R (Ayinde) v The London Borough of Haringey; Hamad Al-Haroun v Qatar National Bank QPSC & Others [2025] EWHC 1383 (Admin) the court said that anyone using AI for legal research has a professional duty to check it against authoritative sources. In Ayinde, the barrister was fined £2,000 and referred to the Bar Standards Board. In the Al-Haroun case, the judge found that Mr Al-Haroun’s and his solicitor’s witness statements relied on 45 authorities, of which 18 did not exist. The solicitors were referred to the SRA.

In May this year, a prominent law firm was heavily criticised by the judge for using an AI hallucinated statute and then for doubling down on the error by relying on AI to try and explain it away rather than admitting the mistake. It had to pay the former client’s costs and referred itself to the Solicitors Regulation Authority (SRA) (Cork v Smith [2026] EWHC 1199 (Ch)).

And, as recently as September, a judge found that an expert’s use of AI in the preparation of the report “crossed the line” as it went beyond helping with presentation and set out conclusions that did not fully accord with the evidence the expert gave in court.  The fact that the AI comments on its own draft were preserved in the report before the court was evidence that there had not been adequate oversight (Cornwall Council v M & Ors [2026] EWFC 306 (30 September 2026)).

There is a growing body of these cases!

To reiterate, Professional Negligence is a failure to meet the standard of a reasonably competent practitioner, causing foreseeable loss.

The Regulators and other bodies have responded:

The SRA issued a warning notice in August 2026 that using AI does not reduce, remove or transfer a solicitor’s professional responsibilities. The notice adds that reliance on AI is not a defence where false material is put before a court.

In September 2026, the Solicitors Disciplinary Tribunal (SDT) struck a lawyer off the register for using misleading AI-generated quotes and citations in defending himself before the SDT (Solicitors Regulation Authority Ltd v Kumar (Case No. 12884-2026)).

Further, the UK Jurisdiction Taskforce – an industry-led initiative created by LawtechUK – concluded in July that existing contract and negligence law can resolve most disputes over AI liability (UK Jurisdiction Taskforce Legal Statement on Liability for AI Harms (July 2026)).

It is clear that professionals may be found negligent for using AI inappropriately, selecting an unsuitable model, failing to carry out appropriate due diligence, or failing to validate AI-generated outputs, including checking for hallucinations. Professionals may also be liable for failing to use AI in circumstances where a competent member of their profession would have done so.

A new danger: Privilege and disclosure

Of particular concern for clients and practitioners is the real and significant danger of losing litigation and professional privilege by uploading advice and documents to public AI tools (such as standard consumer versions of Chat GPT).

In UK v Secretary of State for the Home Department [2026] UKUT 81 (IAC), (another case involving AI hallucinations and false authorities) the Upper Tribunal in immigration proceedings warned that putting client material into public AI tools could breach confidentiality and waive privilege.

Interactions with AI tools do not form a privileged attorney-client relationship, and outputs generated or processed by AI do not inherit legal professional privilege.  This reasoning could mean that inputting confidential information into an open AI tool destroys confidentiality, as disclosure to the AI developer without restrictions on further use could constitute disclosure to the “public”. As such, those communications (such as summarising witness evidence, analysing investigation findings, reviewing legal advice or exploring litigation strategy) and the prompts could become disclosable to the other side.

Practical takeaways: further tips for 2026 and beyond

1. Ask how AI is being used. In 2025, it was recommended that clients enquire what role AI plays in their work. Go further now: ask what checks a human carries out on AI output, and whether client data goes into public tools.
2. Ask who is responsible. It was asked whether a professional could shift responsibility onto the AI provider. The answer is now largely no for regulated professionals. You should still check your engagement letters for exclusions and caps on liability, because contracts are where risk is allocated.
3. Know the AI tool’s limitations. This is still valid. Not all AI systems operate in the same way.
4. Keep up to date with AI regulations. This is still valid.
5. New tip: look in the mirror. If your own business adopts AI-generated statements or uses a customer-facing chatbot, the liability may be yours.
6. New tip: Think before you upload. Uploading documents relating to actual or anticipated litigation, such material to a public AI platform, could put that protection at risk of disclosure to the other side.

Conclusion

Two years ago, the question was “who is to blame when the machine gets it wrong.” The answer has turned out to be reassuringly old-fashioned: the person you paid for the advice.

But a new danger has come to the fore: uploading documents and advice to open AI platforms may expose them, and the prompts, to disclosure to the other side.

If you would like further information and advice about use of AI in professional services, please contact one of our Professional Negligence solicitors by email, or call 03333 231 580.