Skip to content
Back to Errors

The forwarded answer

An AI answer passed to a partner, counsel or client without anyone reading or checking it on the way.

  • First recorded: June 2023
  • Where it shows up: emails to partners · instructions to counsel · witness statements · client updates
  • Status: under enforcement

The short version

An answer from an AI tool is sent on to a partner, to counsel or to a client with a covering line and no reading in between. If the recipient then asks which source supports a point, all the sender has to show is the tool's word for it, and it is the sender, not the tool, who answers for what was sent.

What it is

The forwarded answer is delegation with the checking step left out. It runs in every direction. A trainee sends a partner what the tool produced. A solicitor puts a client's research into a document for the court. A partner signs what was put in front of them. Each time the work changes hands, and nobody between the tool and the reader opens a source.

The Divisional Court dealt with this in R (Ayinde) v London Borough of Haringey and Al-Haroun v Qatar National Bank [2025] EWHC 1383 (Admin), judgment of 6 June 2025. Lawyers who use AI for legal research "have a professional duty therefore to check the accuracy of such research by reference to authoritative sources, before using it in the course of their professional work" (paragraph 7). That duty, the court said, rests on lawyers who do the research themselves "or rely on the work of others who have done so", and it is "no different from the responsibility of a lawyer who relies on the work of a trainee solicitor or a pupil barrister" (paragraph 8).

Read the judgment on the National Archives: caselaw.nationalarchives.gov.uk/ewhc/admin/2025/1383.

A sighting

In Al-Haroun, the schedule of references before the court listed forty-five citations. In eighteen instances the case cited did not exist, and others did not contain the quotations attributed to them or did not support the propositions for which they were cited (paragraph 74). The client accepted that the citations had been generated using publicly available AI tools, legal search engines and online sources (paragraph 76). The solicitor had used the client's research.

The court called it "extraordinary that the lawyer was relying on the client for the accuracy of their legal research, rather than the other way around" (paragraph 79). It said that "A lawyer is not entitled to rely on their lay client for the accuracy of citations of authority or quotations that are contained in documents put before the court by the lawyer", found "a lamentable failure to comply with the basic requirement to check the accuracy of material that is put before the court", and referred the solicitor to the Solicitors Regulation Authority (paragraph 81). In the same paragraph it said it was satisfied that the solicitor "did not realise the true position".

The same pattern appears one step removed in the Ayinde half of the judgment. The law centre that instructed counsel said in evidence that it had not been its practice "to verify the accuracy of case citations or to check the genuineness of authorities relied on by counsel" (paragraph 59).

A lay version of it reached the First-tier Tribunal in Harber v Commissioners for HMRC [2023] UKFTT 1007 (TC). The appellant, who represented herself, relied on nine tribunal decisions that she said had come from "a friend in a solicitor's office" (paragraph 15). The Tribunal found that they were not genuine and had been generated by an AI system such as ChatGPT (paragraph 21), and that she had not known they were fabricated (paragraph 22). Read it at caselaw.nationalarchives.gov.uk/ukftt/tc/2023/1007.

Why it happens

The answer arrives formatted, fluent and complete, so it looks like finished work, and passing work on is an ordinary part of practice. The difference is in what can be asked afterwards. A trainee can be asked where an authority came from and can show the report. A tool can be asked the same question, but its answer is more text from the same tool, not a source (see "All citations verified.").

What to say back

"Which of these points has someone read in the source, and which came straight from the tool? Anything that came straight from the tool has not been checked yet, whoever's name is on the email."

Origin

  • 22 June 2023. In Mata v Avianca, a United States case, one lawyer prepared an affidavit that attached fake decisions and another signed it. The court recorded the account that the signing lawyer "looked it over, and he signed it", and found no evidence that he "asked a single question" (Opinion and Order on Sanctions, Findings of Fact, paragraph 21).
  • 4 December 2023. Harber: a litigant in person relies on AI-generated decisions passed to her by a third party.
  • 6 June 2025. Ayinde and Al-Haroun: the Divisional Court states that the duty to check covers lawyers who rely on the work of others (paragraph 8), and refers a solicitor who relied on his client's research to the SRA (paragraph 81).
  • Now. Paragraph 23 of Ayinde lists what a court can do when the duty is not met: public admonition, a costs order, a wasted costs order, striking out, referral to a regulator, contempt proceedings and referral to the police.

Sources

  1. R (Ayinde) v London Borough of Haringey and Al-Haroun v Qatar National Bank [2025] EWHC 1383 (Admin),
  2. Harber v Commissioners for HMRC [2023] UKFTT 1007 (TC),
  3. Mata v Avianca, Inc., 22-cv-1461 (PKC) (S.D.N.Y.), Opinion and Order on Sanctions,