*By Charlotte Gifford, Senior Money Reporter, 20 Aug 2025 – 02:42PM BST*
A taxpayer has been scolded by a judge for using an Artificial Intelligence (AI) chatbot to fight his £13,000 tax case.
Marc Gunnarsson appealed after HM Revenue and Customs (HMRC) chased him for self-employment support payments totalling £12,918, which he had claimed during the pandemic.
Mr Gunnarsson, who represented himself at the Upper Tribunal, used AI to help with his written submissions, but the chatbot “hallucinated” non-existent cases.
HMRC noticed the three fictitious tribunal decisions when reading through his skeleton argument, which was submitted the day before the hearing.
**‘A serious matter’**
Judge Rupert Jones said: “The accuracy of AI should not be relied upon without checking, particularly when it comes to statements or arguments that it makes concerning the law. There is a danger that unarguable submissions or inaccurate or even fictitious information or references may be generated.”
Judge Jones also said Mr Gunnarsson was not “highly culpable” because he was not legally trained and “may not have understood that the information and submissions presented were not simply unreliable but fictitious.”
However, the judge warned that “in the appropriate case, the Upper Tribunal may take such matters very seriously.”
Mr Gunnarsson had claimed two coronavirus support payments under the Self-Employment Income Support Scheme but HMRC later decided he was not eligible for either of them. Mr Gunnarsson, who had been receiving employment income as a company director, said it had been his honest belief that he was self-employed and therefore eligible for the payments.
The First Tier Tribunal (FTT) allowed Mr Gunnarsson’s claim. However, HMRC then appealed to the Upper Tribunal which determined that the FTT had erred in its decision and HMRC was permitted to recover the money.
**‘Hallucinated’ evidence**
Litigants are increasingly using artificial intelligence to back up their arguments in legal proceedings.
However, AI can sometimes conjure up false information and present it as accurate – called a “hallucination”.
A junior barrister, Sarah Forey of 3 Bolt Court Chambers. was recently accused of using AI after citing fictional cases in her defence of a homeless man at the High Court.
The court did not confirm whether AI was used; however, the judge said it was “improper”, “unreasonable” and “negligent” that she had presented fake cases to the court and ordered her and the solicitors who instructed her to pay wasted costs.
In another case heard this year – Bodrul Zzaman vs The Commissioners for HMRC – a father appealing a £2,500 High Income Child Benefit Charge used AI to construct his defence at the First Tier Tribunal. But the AI spouted out irrelevant arguments and his appeal was dismissed.
The judge said the case “highlights the dangers of reliance on AI tools without human checks to confirm that assertions the tool is generating are accurate.”
lukehebb on
More cases like this might result in people learning what LLMs actually are. Just really advanced predictive text.
Glittering_Copy8907 on
This crap is everywhere at the minute. See it all over social media and Reddit, people using LLM crap to “prove” things relating to legislation and it’s just so so wrong so much of the time but they post it verbatim as if they’ve cited a legal text.
And they won’t listen to you cus GPT told them so and so it must be true
chronicnerv on
The biggest threat to the legal system in the future is how quickly AI can be used to identify and exploit the loopholes that already allow the wealthy to bypass accountability.
At the same time, there’s now a tool that ordinary people can use to challenge authority and consume their time in the same way large corporations and powerful firms have long done to individuals who can’t afford lengthy litigation.
It’s not just about accuracy—AI may soon be better than judges at interpreting the law. The real issue is its ability to expose and disrupt the systemic advantages that the wealthy currently enjoy.
Express-Doughnut-562 on
It’s crap now, but in the future it’s entirely possible that specialised LLMs would get really good at this sort of stuff – far better than humans. Have to say huge chunks of the legal profession are under threat.
Misty_Pix on
Lol ye AI can hallucinate but any smart person checks the sources 🤣
VettelS on
>Mr Gunnarsson, who represented himself at the Upper Tribunal
And why was he representing himself?
Because, since 2013, legal aid is not available for these kind of cases. So people are forced to represent themselves against the professional legal representation of HMRC, in the likely event that they’re unable to afford the thousands or tens of thousands of pounds that private representation would cost.
This is not only unjust, but also may nullify the purported cost savings of cutting legal aid in the first place.
We now have a system where the state is able to take you to court, refuse you professional legal representation, whilst of course availing themselves of that same opportunity. It makes a mockery of the idea of a fair trial when the balance of power is tilted so far in favour of the state, as it prevents the tribunal from reaching an impartial decision and risks unjust outcomes.
Ironically, LLMs *could* be more useful in tribunals and trials of all kinds, but the Government also fails to properly publish legislation and case law. Lawyers are able to access private (and very expensive) law archives, but these are not publicly available, and the resources that are (such as BAILII) are not comprehensive, not up-to-date, and lack the context and analysis that the average layperson requires in order to interpret the relevance of a given case and the legal jargon contained within it.
Again, this makes a mockery of the idea of “open justice” – especially in a common law setting where the law itself is primarily generated *though case law* (i.e. judicial decisions made at trial). Without proper access to all of this, the layperson can *never* have a proper understanding of how they stand in the eyes of the law, and this is plainly a ludicrous situation, which is made worse when the professional lawyers who *do* have access to this information are financially unavailable to them. LLMs are progressing quickly, but no LLM will ever be able to give accurate answers if the underlying information is inaccurate or incomplete.
Sea-Caterpillar-255 on
If you refuse to give people legal representation they will get what they can where they can. This is the false economy of cutting legal representation.
iamezekiel1_14 on
Surely this is a slam dunk case of perjury here? 100% deserved as well.
TapKey4446 on
South Park really nailed their latest episode with this crap
10 commenti
Article contents:
*By Charlotte Gifford, Senior Money Reporter, 20 Aug 2025 – 02:42PM BST*
A taxpayer has been scolded by a judge for using an Artificial Intelligence (AI) chatbot to fight his £13,000 tax case.
Marc Gunnarsson appealed after HM Revenue and Customs (HMRC) chased him for self-employment support payments totalling £12,918, which he had claimed during the pandemic.
Mr Gunnarsson, who represented himself at the Upper Tribunal, used AI to help with his written submissions, but the chatbot “hallucinated” non-existent cases.
HMRC noticed the three fictitious tribunal decisions when reading through his skeleton argument, which was submitted the day before the hearing.
**‘A serious matter’**
Judge Rupert Jones said: “The accuracy of AI should not be relied upon without checking, particularly when it comes to statements or arguments that it makes concerning the law. There is a danger that unarguable submissions or inaccurate or even fictitious information or references may be generated.”
Judge Jones also said Mr Gunnarsson was not “highly culpable” because he was not legally trained and “may not have understood that the information and submissions presented were not simply unreliable but fictitious.”
However, the judge warned that “in the appropriate case, the Upper Tribunal may take such matters very seriously.”
Mr Gunnarsson had claimed two coronavirus support payments under the Self-Employment Income Support Scheme but HMRC later decided he was not eligible for either of them. Mr Gunnarsson, who had been receiving employment income as a company director, said it had been his honest belief that he was self-employed and therefore eligible for the payments.
The First Tier Tribunal (FTT) allowed Mr Gunnarsson’s claim. However, HMRC then appealed to the Upper Tribunal which determined that the FTT had erred in its decision and HMRC was permitted to recover the money.
**‘Hallucinated’ evidence**
Litigants are increasingly using artificial intelligence to back up their arguments in legal proceedings.
However, AI can sometimes conjure up false information and present it as accurate – called a “hallucination”.
A junior barrister, Sarah Forey of 3 Bolt Court Chambers. was recently accused of using AI after citing fictional cases in her defence of a homeless man at the High Court.
The court did not confirm whether AI was used; however, the judge said it was “improper”, “unreasonable” and “negligent” that she had presented fake cases to the court and ordered her and the solicitors who instructed her to pay wasted costs.
In another case heard this year – Bodrul Zzaman vs The Commissioners for HMRC – a father appealing a £2,500 High Income Child Benefit Charge used AI to construct his defence at the First Tier Tribunal. But the AI spouted out irrelevant arguments and his appeal was dismissed.
The judge said the case “highlights the dangers of reliance on AI tools without human checks to confirm that assertions the tool is generating are accurate.”
More cases like this might result in people learning what LLMs actually are. Just really advanced predictive text.
This crap is everywhere at the minute. See it all over social media and Reddit, people using LLM crap to “prove” things relating to legislation and it’s just so so wrong so much of the time but they post it verbatim as if they’ve cited a legal text.
And they won’t listen to you cus GPT told them so and so it must be true
The biggest threat to the legal system in the future is how quickly AI can be used to identify and exploit the loopholes that already allow the wealthy to bypass accountability.
At the same time, there’s now a tool that ordinary people can use to challenge authority and consume their time in the same way large corporations and powerful firms have long done to individuals who can’t afford lengthy litigation.
It’s not just about accuracy—AI may soon be better than judges at interpreting the law. The real issue is its ability to expose and disrupt the systemic advantages that the wealthy currently enjoy.
It’s crap now, but in the future it’s entirely possible that specialised LLMs would get really good at this sort of stuff – far better than humans. Have to say huge chunks of the legal profession are under threat.
Lol ye AI can hallucinate but any smart person checks the sources 🤣
>Mr Gunnarsson, who represented himself at the Upper Tribunal
And why was he representing himself?
Because, since 2013, legal aid is not available for these kind of cases. So people are forced to represent themselves against the professional legal representation of HMRC, in the likely event that they’re unable to afford the thousands or tens of thousands of pounds that private representation would cost.
This is not only unjust, but also may nullify the purported cost savings of cutting legal aid in the first place.
We now have a system where the state is able to take you to court, refuse you professional legal representation, whilst of course availing themselves of that same opportunity. It makes a mockery of the idea of a fair trial when the balance of power is tilted so far in favour of the state, as it prevents the tribunal from reaching an impartial decision and risks unjust outcomes.
Ironically, LLMs *could* be more useful in tribunals and trials of all kinds, but the Government also fails to properly publish legislation and case law. Lawyers are able to access private (and very expensive) law archives, but these are not publicly available, and the resources that are (such as BAILII) are not comprehensive, not up-to-date, and lack the context and analysis that the average layperson requires in order to interpret the relevance of a given case and the legal jargon contained within it.
Again, this makes a mockery of the idea of “open justice” – especially in a common law setting where the law itself is primarily generated *though case law* (i.e. judicial decisions made at trial). Without proper access to all of this, the layperson can *never* have a proper understanding of how they stand in the eyes of the law, and this is plainly a ludicrous situation, which is made worse when the professional lawyers who *do* have access to this information are financially unavailable to them. LLMs are progressing quickly, but no LLM will ever be able to give accurate answers if the underlying information is inaccurate or incomplete.
If you refuse to give people legal representation they will get what they can where they can. This is the false economy of cutting legal representation.
Surely this is a slam dunk case of perjury here? 100% deserved as well.
South Park really nailed their latest episode with this crap