Industries / Finance and business services / Legal and professional services
Courts and filings
How courts receive, store and publish case papers and hearings, and how people reach them.
4
Milestones achieved, from 1988 to 2026.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1988
Court records online
The Judicial Conference of the United States set up PACER, Public Access to Court Electronic Records, in 1988. It now gives access to more than 1 billion documents filed at more than 200 federal courts and is paid for by user fees.
Credit
Judicial Conference of the United States
2016
A court that is online
British Columbia's Civil Resolution Tribunal took its first disputes, over strata (condominium) property, on 13 July 2016, and calls itself Canada's first online tribunal. It now also hears small claims of $5,000 or less and motor vehicle injury disputes, and it received 59,740 claim applications up to 31 March 2026.
Credit
Civil Resolution Tribunal, Province of British Columbia
2025
Sealed papers back on paper
After cyberattacks aimed at the judiciary, the federal court for the Eastern District of Missouri ordered that from 18 August 2025, highly sensitive sealed documents in criminal cases be handed in on paper instead of being filed in its electronic case system, CM/ECF.
LessonA filing system that holds sealed records is a target and has to be secured like one.
Credit
US District Court for the Eastern District of Missouri
2026
2,306 rulings on fake citations
A public database kept by the researcher Damien Charlotin lists 2,306 court and tribunal decisions, as of 16 November 2026, about AI-generated content such as invented citations in filings. 1,566 are from the United States, followed by Canada (245), Australia (112) and the UK (91).
Credit
Damien Charlotin
Watch for
A new federal filing system
The US federal judiciary approved a fee increase to help pay for replacing its case management and public access systems, at an estimated $700 million to $800 million over six years ending in fiscal year 2030. It plans to store sensitive documents in a secure cloud from 2026 and to deploy the new system in the district courts through 2027. The PACER fee rises from 10 to 12 cents a page on 1 January 2027.
Achieved whenevery district court files and serves its cases on the new system.
Claimed by
Judicial Conference of the United States, June 2026
Open
01
Scoring courtroom transcripts
A public score of speech-to-text tools on court audio, measured against official transcripts, that counts wrong words and also words given to the wrong speaker.
The briefCredit · unclaimed02
Citation screening at filing
A pilot of the check a court clerk's office could run on each brief as it arrives: find every citation, confirm that the case exists and contains the quoted words, and flag the rest for a person to read.
The briefCredit · unclaimed
Open 01 · the brief
Scoring courtroom transcripts
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public score of speech-to-text tools on court audio, measured against official transcripts, that counts wrong words and also words given to the wrong speaker.
Why it is still open
No formal court rule sets an acceptable word error rate, and a transcript can score well on words and still give an answer to the wrong speaker, according to the speech-to-text company Speechmatics (March 2026). People make errors too: in a 2019 study, 27 Philadelphia court reporters, certified at 95% accuracy, wrote down 82.9% of words correctly when the speech was African American English. I found no public test of this kind on US court audio.
Built with what exists
US Supreme Court oral argument audio, posted the same day each argument is heard, and the Court's separately published transcripts; speech-to-text models such as ElevenLabs Scribe v3 and open-weight models; pyannote.audio or another open-source tool that labels who spoke when. (see the Note "Fewer word errors")
First step
Score one tool on five arguments from the current term against the Court's transcripts.
Done looks like
50 hours of argument audio scored for at least four tools, with word error rate and speaker error rate for each tool and for each speaker, and the scoring code published.
Open 02 · the brief
Citation screening at filing
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A pilot of the check a court clerk's office could run on each brief as it arrives: find every citation, confirm that the case exists and contains the quoted words, and flag the rest for a person to read.
Why it is still open
Princeton researchers recommended "technical screening at filing" in May 2026 as a proposal, untested. I found no report of a court screening filed briefs this way.
Built with what exists
Free Law Project's citation lookup API, launched in April 2024 to catch invented citations; eyecite; publicly available federal filings on CourtListener; a language model to compare each quote with the case text.
First step
Run the citation lookup API over 50 public briefs and check every flag by hand.
Done looks like
1,000 briefs from one month of filings in one federal district run through the check, every flag reviewed by a person, and a published report, without naming the filers, giving confirmed errors per 100 briefs and the share of flags that were false.