In the previous post, I briefly explained how gen-AI works and sounded a warning about inappropriate reliance on it. This post shows what happens when lawyers outsource their professional judgement to gen-AI.
Professional legal writing includes advice to clients; arguments to a colleague, third party or opposing counsel; and spoken arguments in court in front of clients, opponents and the judge. To state it bluntly, any professional output where accuracy, attention to detail and judgement are on display. Lawyers’ reputations depend upon these things.
So it would be surprising to find lawyers taking risks here. Nevertheless, examples are regularly reported. The consequences have been serious.
However, as a lawyer, it isn’t just your own professional output you have to be mindful of. You also need to be alert to what others are writing and saying. These professional failures are simply the grown‑up version of the same habit students develop when they rely on gen-AI inappropriately.
Troubling examples
In the first case of its kind that I became aware of, two lawyers were sanctioned by a judge and ordered to pay $5,000 in fines to the court for providing six hallucinated authorities in their brief. The judge found they had acted in bad faith, making “false and misleading statements to the court,” while the lawyers, “continued to stand by the fake opinions after the court and the [other side] questioned whether they existed.”[i]
In ANPV v Secretary of State for the Home Department[ii] there are no less than 35 paragraphs (more than the actual judgment itself) of the judge describing the measures he went to, with counsel, to establish how he came to submit multiple fake authorities in his submissions for the appellant. It took the pair over two hours to go through them in what must have been a most uncomfortable exercise in line-by-line accountability. The judge concluded that he should refer counsel to the Bar Standards Board for wasting the tribunal’s time.
And in Ayinde v London Borough of Haringey,[iii] the judge described “appalling professional misbehaviour” when five citations for the claimant turned out to be fake. When challenged the errant lawyers “tried to finesse them into being ‘minor citation errors’” [70]. The fees and costs orders of counsel and solicitor were reduced by four-figure sums [73ff]. One of the lawyers concerned has since been struck off.[iv]
More recently, the Crown Prosecution Service was found to have included hallucinated authorities in its submissions to the court. This was picked up by opposing counsel before any damage was done. The judge referred to the “serious consequences that an error of this nature might have”, had it not been identified when it was.[v]
When institutional actors make these mistakes, the risks are amplified: errors can affect multiple cases, mislead courts, and undermine public confidence.
I’m not listing these to sensationalise. In fact, there were plenty of additional examples I could have mentioned. What I want to highlight is that the lawyers concerned were held responsible for these errors. In each case, the underlying cause was the same: misplaced trust in a system that produces fluent text, not verified legal reasoning.
Regulators and judiciary: emerging standards and expectations
The judiciary has responded to these instances by publishing guidance.[vi] This is of course directed at judges and those working for them, but the advice has general application to lawyers and students. It frames the use of gen-AI within the broader principle of maintaining the integrity of the administration of justice and the rule of law. While recognising that gen-AI might be useful, it reminds judges that they remain responsible to check the accuracy of any gen-AI output they rely on.[vii]
Importantly it also states baldly that judges “must” read the documents and evidence themselves;[viii] excellent advice that students are well advised to follow. The guidance concludes:
“All legal representatives are responsible for the material they put before the court/tribunal and have a professional obligation to ensure it is accurate and appropriate.”[ix]
The Bar Standards Board warns barristers to be alert to potential errors in AI generated content, and to be “transparent” if they have used gen-AI. “Any inaccuracies introduced through the use of AI or other technologies remain your responsibility.”[x]
Similarly, the Law Society warns solicitors to “maintain effective, professional quality control over their output and use. You should: carefully factcheck its products and authenticate the outputs.”[xi]
Neither the judiciary nor the professional bodies treat using gen-AI as a special category. Misuse manifests as breaches of existing professional obligations: accuracy, diligence, judgement, and transparency.
The consequences of failure can be severe: professional reputation, licence to practice, negligence liability to client, even loss of job.
Time saver or just another tool?
The obvious temptation with gen-AI is that it saves time. I submit that once you have diligently fact-checked its output you will have wasted more time than you saved. And if you take the risk and avoid the fact-checking? Well, you may get away with it. Right up until the time when you don’t.
As a student you may feel immune to these consequences, and perhaps you are. Some of you may not even be planning to become a practicing lawyer. My encouragement to you is to nevertheless get into good habits now, and that starts with reading the source material yourself. Gain confidence and experience in your own ability to research the law while you are in the low-risk university environment.
Lawyering isn’t the only profession where you will be relied upon for your judgment and expertise. If you start outsourcing your [legal] reasoning now, you start outsourcing your future professional identity. And once you start doing that, you lose the very thing the professions values most: your judgement.
Don’t outsource your professional reputation to the machine.
In the next article I’ll turn from professional consequences to academic ones and explain why students often misunderstand what university assignments are actually for.
Subscribe below for updates.
[i] Sara Merken, ‘New York lawyers sanctioned for using fake ChatGPT cases in legal brief (Reuters, 26 June 2023) (Read article)
[ii] ANPV v Secretary of State for the Home Department [2025] Unreported (Read case)
[iii] Ayinde v London Borough of Haringey [2025] EWHC 1040 (Read case)
[iv] Neil Rose, ‘Lawyer struck off for using fake AI-generated cases before SDT’ (LegalFutures, 4 September 2026) (Read article)
[v] Andreea-Maria Tobosaru v Court of Law Craiolva, Romania [2026] EWHC 1720 (Admin), [105] (Read case); Bianco Castro, ‘CPS apologises over AI hallucinations in court documents’ (Law Gazette, 14 July 2026) (Read article)
[vi] Artificial Intelligence (AI) – Judicial Guidance (31 October 2025) (Read guidance)
[vii] Ibid, [3,V]
[viii] Ibid
[ix] Ibid, [3, VI]
[x] The Bar Standards Board Handbook (18 May 2026) , p2 (Read handbook)
[xi] The Law Society, ‘Generative AI – the essentials’ (1 October 2025), [3.1] (Read guidance)
Enjoyed this article?
Read Law Right is a blog about reading effectively, writing persuasively, and developing the understanding, analysis and evaluation skills that help law students to succeed.
Subscribe below to receive notifications of future posts.
Leave a Reply