In February 2026, a dispute over a court-reporting database in England and Wales turned into a much bigger argument about open justice, privacy and what happens when sensitive data meets an AI vendor. The database was Courtsdesk. The government ordered it shut down and its archive deleted. Journalists’ groups called that a blow to open justice. The government said it was protecting victims and defendants whose personal details had been mishandled.
Both things can be partly true, and that’s why this case is worth understanding, especially if you build products on top of public-sector or otherwise sensitive data.
What Courtsdesk was
Courtsdesk was a digital service that gave journalists near real-time access to criminal court data across England and Wales. According to law firm Wiggin’s summary, around 1,500 journalists used it. It pulled scattered information about upcoming hearings and prosecutions into one searchable system with filters and alerts.
That matters because court listings in England and Wales have long been hard to use. Information is spread across individual courts and formats, and reporters, especially at local papers with few staff, often simply don’t know a case is happening until it’s over. Courtsdesk operated under an arrangement with HM Courts and Tribunals Service (HMCTS), the Ministry of Justice and the judiciary to solve exactly that problem.
What the government said happened
HMCTS issued a cessation notice to Courtsdesk in November 2025, citing unauthorised sharing of court data. On 10 February 2026, Courts Minister Sarah Sackman explained the decision in the House of Commons (Hansard record). She said Courtsdesk had been “sharing private, personal and legally sensitive information with a third-party AI company, including potentially the addresses and dates of birth of defendants and victims.”
The government’s position was that this breached the terms under which Courtsdesk received the data, and that the data had to be deleted. The minister also stressed that “no one has deleted any court records”. The official records still exist, and accredited journalists can still access listings through individual courts, the courts and tribunals hearings service and GOV.UK. The Ministry of Justice said it was developing an alternative platform.
What Courtsdesk said
Courtsdesk’s founder, the investigative journalist Enda Leahy, disputed the government’s account. Courtsdesk’s position, as reported at the time, was that the AI company was engaged to test a search feature under a confidentiality agreement and wasn’t allowed to use the data to train models. Leahy also disputed the claim that Courtsdesk had accepted it breached the agreement, and published a detailed rebuttal.
Courtsdesk also argued that the official alternatives aren’t good enough. It released figures, which should be read as Courtsdesk’s own claims, suggesting official listings were accurate on only a small fraction of sitting days and that very large numbers of criminal hearings took place without meaningful advance notice to the press. The National Union of Journalists criticised the deletion order, warning that it would change what information is available to the public.
Why both sides have a point
The open justice argument is simple. Justice is supposed to be seen to be done. If reporters can’t find out when hearings are happening, courts become less visible in practice, even if they’re technically public. A tool that made hearings discoverable was serving that principle, and removing it without a working replacement leaves a gap.
The privacy argument is just as real. Court data includes addresses and dates of birth of defendants, and in some cases victims. Those people didn’t choose to be in the dataset. If that information went to a third party outside the agreed terms, even for testing, that’s a serious data protection issue under UK GDPR, regardless of intent.
The hard part is that “public” and “freely shareable” aren’t the same. A hearing being public doesn’t mean every piece of personal data attached to it can be copied to any vendor.
The lesson for developers and product teams
Set the politics aside and this is a textbook data-governance failure mode, one that’s becoming common as teams rush to add AI features. If you build on sensitive data, here’s what to take from it.
1. Your data licence is your product’s foundation
Courtsdesk’s whole business depended on an agreement with a public body. When that agreement was judged to be breached, the product stopped existing. Know exactly what your data agreements allow, especially around sharing with sub-processors, and treat them as core infrastructure, not paperwork.
2. “Just testing” is still sharing
Sending real personal data to a vendor to evaluate a feature is processing, and often a transfer to a third party. An NDA doesn’t change what the underlying data agreement permits. For evaluations, use synthetic or properly anonymised data, or get explicit written permission first.
3. Minimise before you share
Most AI search or summarisation features don’t need addresses and dates of birth. Strip or pseudonymise identifying fields before data leaves your systems. That reduces risk even when sharing is allowed.
4. Keep a sub-processor register
List every third party that touches the data, what they get, why, and under what contract. If a regulator or partner asks, you should be able to answer in minutes, not weeks.
5. Plan for termination
Ask what happens if your data source withdraws access tomorrow. Can you delete properly and prove it? Do your customers have a fallback? Courtsdesk’s users lost the tool and its archive at the same time.
What happens next
The central question the case raised hasn’t gone away: how should the justice system make court information genuinely accessible to journalists without exposing the personal data of the people involved? A government-run service can work if it’s accurate and timely. Critics worry that a state-controlled portal will be less useful and less independent than a third-party tool built by journalists.
If you follow open justice or build civic-tech products, it’s worth watching how the Ministry of Justice’s replacement service performs, and whether any future arrangement with outside providers includes clearer, stricter rules on AI vendors. The episode is a reminder that the most important decision in an AI feature isn’t the model. It’s which data you let it see.
Related reading
- Who Sees Your Smart Glasses Footage?
- Can Terms of Service Be Updated by Email?
- OpenAI, Anthropic and the Pentagon: A Timeline
Sources: Hansard, 10 February 2026, Wiggin LLP, Legal Cheek, NUJ.
