Britain's courts were transparent 4.2% of the time. The tool that proved it is being deleted
The Ministry of Justice has ordered the destruction of a five-year journalism archive. The stated reason is a data breach. The real story is what the archive revealed about the courts themselves.
Charlie Moloney used to bring scissors to work. The court reporter and media law expert would print out the daily court lists each morning, glue them into a book, then write each defendant's charge beside their name by hand, because HM Courts and Tribunals Service issued two separate documents for the same information and provided no way to cross-reference them. This was not the 1970s. This was the standard method of tracking what happened in England's magistrates' courts well into the last decade.
Then someone built a search engine for it, and the search engine turned up results the Ministry of Justice would rather not have seen.
What Courtsdesk built
Courtsdesk began as a frustration. Enda Leahy, a former senior news correspondent at the Sunday Times in Ireland, had spent years struggling to access court information as an investigative journalist. In 2017, he and co-founder Alan Larkin built a platform to aggregate publicly available data from courts and company registries, initially in Ireland, before expanding to England and Wales. The service was approved as a government pilot in 2020 with the backing of the Lord Chancellor and Justice Minister Chris Philp, and designed to provide a live digital news feed of court listings and registers.
The concept was deceptively simple. HMCTS published court lists across hundreds of magistrates' courts in various formats, on different schedules, through inconsistent channels. Some courts emailed lists to local newsrooms. Others did not. The lists themselves were fragmentary, often separating defendants' names from their charges, making it impossible to identify cases of public interest without physical attendance or the scissors-and-glue method Moloney described. Courtsdesk unified these scattered data streams into a single searchable platform that could deliver up to 12,000 updates in a 24-hour period, filtered by location, offence type, and other criteria. More than 1,500 reporters across 39 media organisations came to rely on it.
The service did what HMCTS had been promising to do for years but had never managed. A 2021 government consultation on open justice acknowledged the problem directly: the Publications and Information project would "digitise the publication of court and tribunal lists of upcoming hearings and make them available online and nationwide." By 2026, that project still had not delivered a functioning national system.
Courtsdesk delivered one in 2020. And then it started counting.
The numbers nobody wanted
Over five years of aggregating court data, Courtsdesk compiled statistics that amounted to a performance audit of the entire magistrates' court system. The findings were remarkable. Court listings matched reality on just 4.2% of sitting days. Approximately 1.6 million criminal hearings proceeded without any advance notice to the press. Half a million weekend cases were heard with no journalist notification whatsoever. Two-thirds of courts routinely heard cases without telling the media they were happening, and some courts never published advance listings at all during the periods measured.
These are not the complaints of a disgruntled contractor. They are measurements. Nat Goodlad, a reporter at the Oldham Times, put it plainly after the deletion order was announced. Courtsdesk, he wrote, "was never a replacement for court lists, phone calls, or sitting through hearings. It was a way of making sense of a system that increasingly withholds detail until the last possible moment." The data did not damage open justice. It demonstrated where open justice was already failing.
The question of why this matters requires a brief detour into what has happened to court journalism in England over the past two decades.
The empty press bench
A 2019 study by researchers at the University of the West of England monitored Bristol Magistrates' Court for an entire week. In that time, 240 cases passed through the building. The researchers spotted a single working journalist. Only three stories from that week appeared in the local press, and the research team identified a significant number of cases they judged newsworthy that went entirely uncovered. "Effectively," the academics concluded, "this local-level justice was being conducted invisibly."
Bristol is not unusual. Academic research by Richard Jones at the University of Huddersfield, based on 22 interviews with court reporters, found that the local daily newspaper correspondent is now typically the only journalist from any news organisation to physically attend a court. Several reported going weeks or months without seeing another reporter. This is particularly acute in magistrates' courts, where all criminal cases begin and the vast majority end, but where media presence has dwindled most severely.
The causes are well documented: declining advertising revenue, staff cuts at local newspapers, the consolidation of titles under corporate publishers servicing pension liabilities rather than investing in journalism. Dame Frances Cairncross's independent report for the government in 2019 identified the collapse in local court coverage as a specific concern. The House of Lords Communications Committee echoed it. Yet HMCTS's own systems remained fragmented and unreliable, and the courts themselves made few structural accommodations for the reporters who remained.
Into this landscape, Courtsdesk functioned as a partial technological remedy for an economic problem. It could not put reporters back on press benches, but it could tell the reporters who remained which benches to sit on. Without it, the selection of which cases to cover reverts to whatever fragments of information happen to reach a newsroom through inconsistent local channels, police press releases issued only after convictions, and word of mouth.
The stated reason and the unstated one
In November 2025, HMCTS issued Courtsdesk a cessation notice, citing "unauthorised sharing" of court data with a third-party AI company. In February 2026, Courts Minister Sarah Sackman told the House of Commons that some 700 individual cases had been shared with an AI company, described the situation as "incredibly serious," and said Courtsdesk had accepted it had breached its agreement.
Leahy's account, set out in detail on Substack the same day, differs sharply. Courtsdesk had hired a specialist firm to build a safety tool for journalists in a secure sandbox environment. The contractors held government security clearance and DBS checks, had previously worked on data systems for the National Archives and the Treasury, and operated under a formal sub-processor agreement prohibiting any use of data beyond providing services to Courtsdesk, any sharing with third parties, and any use for AI model training. The data was processed in an isolated, encrypted sandbox with 24-hour automated deletion via AWS bucket lifecycle rules. Courtsdesk paid the contractor 45,000 pounds a year. No money flowed the other direction.
The minister's characterisation of the arrangement as sharing data with an AI company "no doubt for commercial purposes" was, Leahy argued, a misrepresentation of what was in fact a standard subcontractor relationship, legally classified under UK GDPR as a sub-processor rather than a third party. He also flatly denied that Courtsdesk had accepted any breach, pointing to formal disputes lodged by his lawyers at Bristows and counsel Jude Bunting KC of Doughty Street Chambers.
One detail from the parliamentary exchange stands out above all others. When asked whether the alleged breach had been reported to the Information Commissioner's Office, the regulator responsible for data protection in the United Kingdom, Sackman replied that she had been advised it "did not meet the threshold for a referral."
Leahy seized on this immediately. "You can't have it both ways," he wrote. "You can't call it a serious breach in Parliament and then say your own data protection officer didn't think it was worth reporting."
A familiar pattern
Anyone who followed Meta's shutdown of CrowdTangle in August 2024 will recognise the institutional choreography. CrowdTangle was a social media monitoring tool that allowed researchers, journalists, and civil society groups to track the spread of content across Facebook and Instagram in real time. It was acquired by Meta in 2016 and made freely available. Over the following years, it became indispensable for election integrity research and disinformation tracking, but it also generated data that embarrassed Meta, revealing how the company's algorithms amplified divisive content.
Meta announced CrowdTangle's closure three months before the 2024 US presidential election, pointing users towards a replacement called the Meta Content Library. Researchers immediately flagged that the replacement was less transparent, less accessible, had fewer features, and excluded most journalists from access entirely. A joint investigation by Proof News, the Tow Centre for Digital Journalism, and the Algorithmic Transparency Institute found the new tool inferior on eleven key metrics. Brandon Silverman, CrowdTangle's co-founder, observed that the usability of data is almost as important as the quantity of data itself, and that dumping information onto researchers without functional tools to interrogate it is not meaningful transparency.
The structural parallels are precise. In both cases, a tool was created to make an opaque system legible. In both cases, the tool revealed uncomfortable truths about the institution hosting the data. In both cases, the institution killed the tool while promising a replacement. In both cases, the replacement was described as better while being designed to be more controlled. And in both cases, the institution insisted that the underlying data remained "available" while removing the practical means of using it effectively.
HMCTS's assurance that journalists retain "full access to court information" through existing channels is technically accurate in the same way that Meta's assurance that the Content Library would serve researchers was technically accurate. The access exists. The utility does not.
The deeper tension
Beneath the immediate dispute lies a problem that legal scholars have been wrestling with for two decades: what happens to the principle of open justice when practical obscurity disappears.
The phrase originates in American jurisprudence. In a 1989 case, the US Supreme Court noted that FBI rap sheets enjoyed a degree of "practical obscurity" that would be eliminated if compiled into a single searchable database. Legal scholar David Ardia of the University of North Carolina developed this into a broader framework in 2017, arguing that court records had always contained vast quantities of sensitive personal information, from social security numbers to the names of sexual assault victims, but the privacy harms were historically considered insignificant because the records were so difficult to search and access.
Digitisation changes this calculus entirely. Information that was once practically obscure, requiring a physical visit to a courthouse and a manual search through paper files, becomes instantly accessible, cross-referenceable, and permanent when placed in a searchable electronic database. The same transformation that makes Courtsdesk valuable to journalists makes it potentially dangerous when court data meets AI systems capable of ingesting, correlating, and reproducing personal information at scale.
This is a genuine tension, not a manufactured one. But it is also not a novel one, and other countries have found ways to manage it that do not require choosing between total openness and wholesale deletion. Germany issues a Führungszeugnis, a government certificate that discloses only convictions above a certain severity threshold, shielding minor offences from routine background checks while preserving the full judicial record. France operates a three-tier casier judiciaire: the complete record is accessible only to judges, a second tier is available to certain employers such as those in government or childcare, and citizens themselves can request only a summary showing serious convictions. The Netherlands and Ukraine publish court rulings with personal data redacted, allowing legal research and public scrutiny without exposing individuals to permanent digital stigma. These approaches separate the principle of open justice from the practice of universal searchability. They distinguish between a citizen's right to observe the courts and a corporation's ability to ingest court data for commercial purposes.
The UK has its own version of tiered access in the Disclosure and Barring Service, which operates at four levels of disclosure for employment purposes. But it has no equivalent framework for managing bulk access to court listings and registers in the digital age. Courtsdesk existed in an institutional vacuum, doing work the government should have been doing, measuring performance the government was not measuring, and generating transparency the government had not anticipated.
What happens by the end of March
Minister Sackman told Parliament that all magistrates' court lists and accompanying case summarisation data would be available from a new court and tribunal hearings service by the end of March 2026. She described this as putting the system "on a securer footing with the necessary guardrails." HMCTS is also working on a new licensing arrangement that would allow third parties to apply to use court data under regulated conditions.
These promises may be sincere. They may even be delivered on time. But the institutional track record is not encouraging. HMCTS first announced its Publications and Information project to consolidate court listings in a single digital service in 2021. Five years later, it had not been completed. The Catalogue of Open and Accessible Data promised for 2022 remains incomplete. The media working group established to improve journalist access has met regularly for years without solving the basic problem that Courtsdesk solved in its first months of operation.
There is also a question of what is lost in the interim, and what is lost permanently. Courtsdesk's five-year archive contained not merely the current state of court listings but a longitudinal record of how the system performed over time. Patterns of missing notifications, geographical variation in listing accuracy, weekend hearing volumes, courts that never published advance listings: all of this was measurable because the data existed in aggregate, over years. Delete the archive and the ability to reconstruct that evidence-based narrative disappears. Future claims by HMCTS that performance has improved will be unverifiable against the baseline, because the baseline will have been destroyed.
This is the sharpest edge of the decision. It is one thing to suspend a service pending remediation of a data protection concern. It is another to order the permanent deletion of every record. Suspension preserves the possibility of resolution. Deletion forecloses it.
Open justice in theory and in practice
The principle that justice must be seen to be done is among the oldest in English common law. Lord Chief Justice Hewart's formulation from 1924, that justice "should manifestly and undoubtedly be seen to be done," is cited so frequently it has become almost decorative, invoked in consultation documents and judicial guidance without much examination of whether its conditions are being met.
Courtsdesk provided that examination, and the results were damning. A system in which court listings are accurate less than five per cent of the time, in which more than a million hearings proceed without any journalist knowing they are happening, is not a system in which justice is being seen to be done. It is a system in which justice is being done, or not done, largely in the dark, with occasional shafts of light falling on whatever cases happen to attract the attention of the diminishing number of reporters still working the beat.
The response to this exposure has not been to fix the underlying problem. It has been to shut down the measurement and order the evidence destroyed. The government may well deliver a replacement system by March. It may even be a good one. But the decision to destroy the archive rather than preserve it, to refuse dialogue rather than negotiate safeguards, to cite a data breach too minor for the ICO while treating it as serious enough to justify deletion of five years of journalism infrastructure, suggests that the embarrassment of being measured matters more than the principle of being transparent.
Charlie Moloney no longer needs scissors and glue. But as of this month, he may need them again.