Bandla v Solicitors Regulation Authority

What happened

The appellant, a former practising solicitor struck off the roll in 2017, appealed. The papers were substantial: witness statements in March, April and May 2025, and a bundle running to 1,204 pages. The SRA cited 24 authorities. The SRA’s solicitors also had to prepare a separate bundle containing the 53 cases cited by the appellant which exist, which tells you the scale of the problem before any number is quoted.

The SRA’s annex listed 27 authorities as non-existent. The judge fairly recorded that two of the 27 were wrongly criticised: “Osborne” was Osborn [2013] UKSC 61, and Ghosh was a Privy Council case rather than a House of Lords one.

The exchange that decided it

Put to him on the first example, a citation given as R (on the application of Smith) v Parole Board, the appellant said he had not written the summary himself and had not read the judgment himself. He denied using AI or any source identifiable as AI, and said he had simply used a Google search for “case law in support of mental health problems”.

He accepted the cases did not exist. He said he never “double-verified” them, and then accepted that he never checked them at all.

Asked why the court should not strike out the grounds of its own motion, he argued the substantive points were sound even if the authorities cited for them did not exist. The judge was, in his own word, wholly unpersuaded.

The holding

“In my judgment, the Court needs to take decisive action to protect the integrity of its processes against any citation of fake authority.”

Note the words “any citation of fake authority”. The rule is tool-agnostic, which is what makes this judgment more durable than the ones that turn on a finding about ChatGPT. The court added weight to the fact that the citations came from someone who had previously been a practising solicitor, and that they were maintained after the SRA had pointed them out well ahead of the hearing.

• ExtensionApplication for an extension of time dismissed.
• Strike outGrounds of appeal struck out in any event, as an abuse of the process of the court.
• CostsThe SRA applied for costs on an indemnity basis, summarily assessed.

What it changes

Two contributions. First, it puts the English courts alongside Connecticut on the question of who pays. Connecticut’s Practice Book section 4-9 names nonsuit and default judgment; Bandla is what that looks like when a court actually does it. Second, it is the answer to “I did not use AI”. Together with Whiting and In re Nwaubani, it establishes that a denial does not stop the consequence, because the consequence attaches to the fake citation, not to the tool.

Sources

• Judgment, [2025] EWHC 1167 (Admin), Find Case Law, The National Archives, read in full
• Cited in the judgment: Osborn v Parole Board [2013] UKSC 61

Related

• The full sanctions record
• Ayinde v Haringey, the leading England and Wales authority
• Harber v HMRC, the earliest UK instance
• England and Wales: the judiciary’s AI guidance

Checked 5 August 2026 against the judgment as published by The National Archives. The court made no finding that AI produced the citations, and we say so wherever we describe this case.

Common questions

What was decided in Bandla v SRA?

The court dismissed an application for an extension of time and, in any event, struck out the grounds of appeal as an abuse of the process of the court, because the appellant had cited non-existent authorities and maintained them. The SRA applied for costs on an indemnity basis.

How many fake cases were there in Bandla?

The SRA's annex listed 27 authorities as non-existent. The judge recorded that two of the 27 were, in his view, wrongly criticised: one was a real case whose name had been misspelled, and one was a real case attributed to the wrong court. The appellant accepted that the cases did not exist and that he had never checked them at all.

Did the appellant admit using AI?

No. He denied using AI or any source identifiable as AI, and claimed to have simply used a Google search for 'case law in support of mental health problems'. He also said he had not written the case summary himself and had not read the judgment himself.

Why does Bandla matter if AI was denied?

Because the court's reasoning does not depend on the tool. It held that the court needs to take decisive action to protect the integrity of its processes against any citation of fake authority. That rule applies whatever produced it.

Is striking out an appeal a normal sanction for fake citations?

It is unusual, and it is the heaviest form on this record short of criminal proceedings, because the loss falls on the case rather than on the lawyer's wallet. It is the English analogue of Connecticut Practice Book section 4-9, which names nonsuit and default judgment among its available sanctions.