Skip to content

reportAI in CourtsAI Governance

Where justice systems use AI: 595 official records as of 7 September 2026

595 official justice-sector AI records in 17 jurisdiction codes: 351 DOJ rows, 98 Brazil rows and 31 Netherlands rows as of 7 September 2026.

Edited and verified by Cognesio LLP

Researched with AI assistance · sources verified by Cognesio LLP · How this was made ↓

As of 7 September 2026, the SafeLegalAI public-sector register records 595 official records in 17 country, territory or EU codes for AI used, piloted or procured by justice bodies: courts, judiciaries, prosecutors, public defenders, legal-aid bodies, tribunals, justice ministries, corrections agencies, police-forensics bodies and court administrators. The United States supplies 412 rows, including 351 rows from Department of Justice AI inventories; the United Kingdom supplies 21 rows from Algorithmic Transparency Records, Ministry of Justice announcements, court announcements, HMPPS records, Cafcass records and prosecution policy.

The record is not a count of every system inside justice. It is a count of public records. It shows which bodies chose inventory entries, transparency records, procurement notices, official announcements, pilot reports, judgements or policies as the public form for saying what AI they use, test, plan or have stopped.

Key findings

  1. On 7 September 2026 the official record held 595 entries from 17 country, territory or EU codes; 585 were read from the official record, 10 carried an official link not yet read, and the read set came from 100 primary source URLs.
  2. The United States supplied 412 of the 595 justice-sector records on 7 September 2026, including 351 Department of Justice rows collected from the 2024 and 2025 AI inventories; the live 2025 DOJ spreadsheet contained 314 use-case rows, while this release held 301 DOJ rows dated 2025.
  3. The DOJ rows in the register contained 92 generative-AI rows on 7 September 2026, which is 26.2 percent of the 351 DOJ rows; the dataset status words were 155 “Deployed”, 103 “Pre-deployment”, 25 “Pilot”, 26 “Retired”, 32 “Operation and Maintenance”, 6 “Acquisition and/or Development” and 4 “Initiated”.
  4. Brazil supplied the second-largest national block on 7 September 2026, with 98 justice-system records: one CNJ policy, five announcements and 92 CNJ survey entries from 56 tribunals.
  5. The Netherlands supplied 31 justice-family algorithm-register rows on 7 September 2026; none was coded generative AI, while 13 were analytics and 10 were risk assessment.
  6. The United Kingdom supplied 21 justice-family records on 7 September 2026, including 9 Algorithmic Transparency Records, 9 official announcements, 2 pilot reports and 1 prosecution policy.
  7. Four official court records disclosed judicial AI use by 20 February 2025: Snell, Deleon, Ross v United States and Colombia’s T-323/24.
  8. Judicial and justice-body governance appeared in 27 policy or judicial-conduct records on 7 September 2026, including US federal guidance, 18 US state or local policy rows, two US judicial-conduct opinions and six non-US policy rows.

Why this question

Justice-sector AI is often described through rules: what judges, lawyers or court staff may put into a tool, what they must verify, and whether they must tell parties. Those rules matter, but they are only half the official record. The other half is the public inventory: named systems, status labels and use cases.

The labels matter.

This report answers a narrower question than the policy debate. It asks what official inventories, transparency registers, procurement portals, court announcements and judgements say justice bodies actually use, pilot, plan, procure or retire as of 7 September 2026. A separate SafeLegalAI report covers the rules that bind judges’ AI use and the adjudicator hallucination cases: the bench’s record. This report does not repeat that analysis. It follows the deployment record.

The question is practical. A ministry policy may say a model may not decide a case. A public inventory says whether the same ministry has a tool for triage, transcription, redaction, identity matching, searching prison reports, drafting minutes, summarising case files or supporting court listing. A contract award says money has been committed. A pilot report says a test happened. A judgement disclosure says the court used a model in a live opinion. Those are different records, and this dataset keeps them apart.

Method and data

The dataset is the SafeLegalAI public-sector register, browsable at /public-sector, by body at pages such as /public-sector/body/us-doj-dea, by country at pages such as /public-sector/country/us, exported at /public-sector.json, and released on GitHub at github.com/SafeLegalAI/justice-ai-deployments. Each row records the body, country, record type, tool or use case, AI kind, source status phrase, date, any published money figure, summary, quote, licence, judge name where printed, source URL, archive URL and verification note.

Every number in this report is computed by the Python script reproduced in the run log. Body type, record type and AI kind use the labels on the public-sector pages. Country names use the site country map. The dataset manifest records 595 rows and no rejected rows.

Status words are the source’s own status words on the read date. They are never upgraded. A pilot is not a deployment. A contract or procurement notice is not use. A row whose source says “to be developed and tested” is reported that way, even where later work is expected.

The DOJ 2024 and 2025 inventories were both read. The live 2025 DOJ spreadsheet contained 314 public use-case rows, with 163 “Deployed”, 107 “Pre-deployment”, 25 “Pilot” and 19 “Retired”. This register held the rows collected on 7 September 2026: 50 DOJ rows dated 2024 and 301 dated 2025. It does not claim to be the full live spreadsheet. Matching on component and use-case name found one exact repeat, ProLaw in the Professional Responsibility Advisory Office, so the two dated source entries are retained; product names used by different components remain separate component records.

Published money figures are kept as published and are never summed across currencies. The dataset has three TED rows with a published amount, shown in Appendix A only. The TED notices were readable when the register was built; a later fact-check met a WAF challenge and found no Wayback capture for the three TED notice URLs.

Licences remain source-specific. US federal inventory rows are treated as US federal public-domain material; US court opinions are public-domain court opinions; UK government material is recorded under OGL v3 where the source carries that status; EU procurement notices carry EU reuse terms; Brazil, Colombia, China and several judiciary policies are official acts or quoted-short records. The dataset’s SafeLegalAI coding is CC BY 4.0. Ten rows have an official link recorded but are not quoted as read primary text. There are no rows read from public mirrors in this public-sector dataset.

CEPEJ, the National Center for State Courts and the AI Incident Database were lead sources only. No table below is counted from those sources.

Verification statusRows
Read from the official record585
Official link recorded, text not quoted10
Read from a public mirror0

The DOJ inventory is the largest block

The United States DOJ rows make the register look different from the public-facing court record: most rows are not judge-facing tools, but component inventory entries for investigations, prisons, prosecution, administration, immigration adjudication and department-wide services. The DOJ register subset has 351 rows, 92 generative-AI rows and seven dataset status words. One table drives the section.

DOJ componentRowsGenerative AI
Drug Enforcement Administration5518
Federal Bureau of Investigation4811
Bureau of Alcohol, Tobacco, Firearms and Explosives393
Federal Bureau of Prisons386
Department-wide203
United States Marshals Service187
Civil Division173
Tax Division144
Antitrust Division107
Justice Management Division / Organized Crime Drug Enforcement Task Forces104
All other DOJ components8226

By body type, the DOJ rows are concentrated in police and forensics: 164 of 351. Prosecutors account for 66 rows, other justice agencies for 48, corrections and probation for 45, justice ministry rows for 20 and tribunal or adjudicator rows for 8. The 8 tribunal rows are the Executive Office for Immigration Review.

DOJ body typeRows
Police and forensics164
Prosecution service66
Other justice agency48
Corrections and probation45
Justice ministry20
Tribunal or adjudicator8

The DOJ AI-kind distribution is not a generative-AI-only record. Generative AI is the largest single kind, but it is 92 of 351 rows, or 26.2 percent. Document automation has 55 rows, analytics 53, search and retrieval 46, risk assessment 24, transcription 18, NLP / classification 18, not specified 17, translation 12, other 7, chatbot 6 and record linkage 3.

DOJ AI kindRows
Generative AI92
Document automation55
Analytics53
Search and retrieval46
Risk assessment24
Transcription18
NLP / classification18
Not specified17
Translation12
Other7
Chatbot6
Record linkage3

The source status words show a working inventory, a pipeline and a retirement list. DOJ supplies 155 “Deployed” rows and 103 “Pre-deployment” rows. It also supplies 25 “Pilot” rows and 26 “Retired” rows. The remaining DOJ status words are “Operation and Maintenance” for 32 rows, “Acquisition and/or Development” for 6 and “Initiated” for 4.

DOJ status as statedRows
Deployed155
Pre-deployment103
Operation and Maintenance32
Retired26
Pilot25
Acquisition and/or Development6
Initiated4

The named use cases show what those categories mean in the department’s own words. The FBI writing assistant, sourced to the DOJ AI inventory, is pre-deployment and generative AI; the public-domain inventory quote is “Quality and consistency of written comments”. The ATF Veritone Redact row is deployed and generative AI; the official sentence says, “The purpose of the AI use case in Veritone Redact is to support ATF in processing Freedom of Information Act (FOIA) requests by automating the redaction of sensitive information within audio and video files.”

The Office of the Pardon Attorney voicemail row is a pilot, not a deployment. Its official sentence says, “PARDON would like to leverage transcription, translation and summarization services available in the GCC high cloud environment to help reduce the processing time and level of effort associated with responding to voicemail inquiries.” The department-wide Westlaw AI-assisted legal research row is deployed and search/retrieval; the official quote says it “Addresses manual process of conducting legal research.”

Other DOJ rows are plainly operational but not legal drafting. The USMS video transcription tool is deployed transcription; the quote says USMS uses open-source NLP and Python code to transcribe audio or video into text. The Bureau of Prisons Wellsaid row is deployed generative AI; the inventory says, “AI converts text to speech audio that is used in social media, training videos, and audiobooks.” These are official records of use, not claims about outcomes.

The UK record mixes transparency records, ministry announcements and court work

The United Kingdom set has 21 rows. Nine are Algorithmic Transparency Records, nine are official announcements, two are pilot reports and one is a prosecution policy. The set includes MoJ, OPG, HMPPS, HMCTS, the Probation Service, Cafcass, CPS, COPFS and the Courts and Tribunals Judiciary. It is a family record, not one inventory.

UK record typeRows
Transparency record9
Official announcement9
Pilot report2
Internal policy1

The 2026 rows show pre-deployment, equipped staff, tests and studies. The OPG Investigations AI Assistant row is an Algorithmic Transparency Record dated 7 August 2026 and has the status “pre-deployment”; the official record says the assistant is for investigations work. The Justice Transcribe announcement records that “every probation officer in England and Wales has been equipped with Justice Transcribe” and is sourced to the 8 June 2026 justice tech announcement.

The same 8 June 2026 announcement supplies two other rows. The AI legal assistants row is generative AI with status “to be developed and tested”; it is not counted as deployed. The HMCTS trial-listing support tool row is analytics with status “new AI tool”. The AI court transcription study, from a 13 April 2026 HMCTS announcement, is a study and is recorded as “new study will explore”.

The 2025 UK records show the difference between an ATR and a programme announcement. The nDelius contact log semantic search row is an ATR with “public-beta” status and the official ATR says it helps probation practitioners find information within offender reports. The Splink master record and Data First Splink rows are record-linkage records, both linked to official transparency records.

HMPPS supplies a production risk-assessment row, a production NLP/classification row, an announcement about scanned prison phones and a pilot report on prison education content. The Violence in Prisons Estimator row is a production Algorithmic Transparency Record. The Key Work quality assurance automation row is also a production ATR. The digital education content pilots row is a pilot report and remains coded as a pilot report.

UK rows that name courts or judges are sparse. The Courts and Tribunals Judiciary Microsoft 365 Copilot row, sourced to the AI Action Plan for Justice, records leadership judges as a rollout group. The HMCTS intelligent document processing pilot concerns paper forms, not adjudication. The COPFS policy is a prosecution policy; the official page says, “AI will support our work, but it won’t make decisions or replace the judgement of our skilled people.”

Selected UK rowsBodyStatus as statedAI kind
OPG Investigations AI AssistantOffice of the Public Guardianpre-deploymentDocument automation
Justice TranscribeThe Probation ServiceequippedTranscription
AI legal assistants for justice staff and professionalsMinistry of Justiceto be developed and testedGenerative AI
AI trial-listing support toolHMCTSnew AI toolAnalytics
nDelius contact log semantic searchMinistry of Justicepublic-betaSearch and retrieval
Correspondence Drafting ToolCrown Prosecution Servicebeta-pilotGenerative AI
Violence in Prisons EstimatorHMPPSproductionRisk assessment
Cafcass Genesys chatbotCafcassproductionChatbot

Brazil and the Netherlands show two register models

Brazil and the Netherlands together provide 129 rows, but the public records are different. Brazil’s block is a CNJ survey of tribunal projects, plus CNJ policy and court announcements. The Netherlands block is a national algorithm-register slice across justice, police, child-protection, legal-aid and forensic bodies. Both keep source status labels.

Brazil has 98 rows in the register. The CNJ survey subset has 92 entries from 56 tribunals. It records 56 generative-AI rows, 17 search/retrieval rows, 12 NLP/classification rows, 3 transcription rows, 2 chatbot rows, 1 analytics row and 1 risk-assessment row. It also records project stage in Portuguese.

Brazil CNJ survey status as statedRows
Finalizado e em produção / uso38
Em desenvolvimento25
Em teste14
Finalizado, mas não está em produção6
Concepção (ideação e planejamento)5
Finalizado, aguardando autorização para entrar em produção / uso4

The CNJ rows should not be recoded into English maturity stages. “Finalizado e em produção / uso” is counted as the source phrase. “Em desenvolvimento” is not turned into production. The CNJ examples include DEGRAVA at TRT8, a transcription row in production/use; GALILEU at TRT4, a generative-AI row in production/use; and GERADOR DE EMENTAS TJSP, a TJSP generative-AI row in production/use.

The policy row is separate. CNJ Resolution 615/2025, linked to the CNJ official PDF, is a policy row with the source status phrase “RESOLVE”. That is a governance record, not one of the 92 project entries.

Brazil CNJ survey AI kindRows
Generative AI56
Search and retrieval17
NLP / classification12
Transcription3
Chatbot2
Analytics1
Risk assessment1

The Netherlands set is 31 national algorithm-register rows. It contains no generative-AI rows in this release. The largest kinds are analytics (13) and risk assessment (10), followed by search and retrieval (4), document automation (2), NLP/classification (1) and other (1). The main bodies are the Nederlands Forensisch Instituut (7), Politie (6), Raad voor de Kinderbescherming (6), Ministerie van Justitie en Veiligheid (5), Dienst Justitiële Inrichtingen (3), Raad voor Rechtsbijstand (3) and Justis (1).

Netherlands bodyRows
Nederlands Forensisch Instituut7
Politie6
Raad voor de Kinderbescherming6
Ministerie van Justitie en Veiligheid5
Dienst Justitiële Inrichtingen3
Raad voor Rechtsbijstand3
Justis1

The named Netherlands rows show what the national register exposes. Sextortion Classifier is a Politie NLP/classification row. Zoekapplicatie cold cases is search/retrieval and says the algorithm helps find information in cold-case files. Module Zorg is a Raad voor de Kinderbescherming risk-assessment row. Automatische gezichtsbeeldvergelijking is an NFI analytics row for facial image comparison.

Other official records show translation, scheduling, chatbots, procurement and judicial tools

Outside the three largest country blocks and the UK, the register has 33 rows across Estonia, India, Spain, the United Arab Emirates, China, Singapore, Norway, Colombia, Germany, Hong Kong, the European Union, Malaysia and Argentina. The rows are small in count but specific in function: translation, transcription, scheduling, legal chatbots, procurement, summaries and judicial assistants.

CountryRowsSelected record
Estonia5Anonymization of court rulings
India5Supreme Court Portal Assistance in Court Efficiency (SUPACE)
Spain5DELFOS AI search engine for judicial files
United Arab Emirates4Virtual Legal Advisor
China3National-level legal AI judicial platform
Singapore2Generative AI-powered case summarisation tool
Colombia2Copilot for transcription, summaries and act drafting
Norway2IT development services framework for Digital Courts
Germany, Hong Kong, EU, Malaysia, Argentina5One record each

India’s five rows are from Supreme Court and eCourts records. SUPACE is coded search/retrieval with the status phrase “experimental stage of development for its testing”. AI/ML judgement translation records translation from English to 18 Indian languages. AI/ML e-filing defect identification is document automation and says AI and ML tools were developed and deployed inside e-filing. Constitution Bench transcription records transcription used for oral arguments, and eCourts Phase III intelligent scheduling records AI used to recommend scheduling.

Singapore’s two Small Claims Tribunals rows are generative summarisation and translation. The case summarisation row, linked to a Singapore Courts announcement, says the tool summarises case documents for Tribunal Magistrates and self-represented individuals. The document-translation row records generative-AI translations for Chinese, Malay and Tamil court documents.

The UAE’s four Ministry of Justice rows are service descriptions: Virtual Legal Advisor, Virtual Family Advisor, Digital Agencies smart OCR and Ask Adel. They are coded chatbot or document automation. The source phrase is “Service Description”, so the table does not infer usage stage beyond that label.

China’s three rows include the Supreme People’s Court legal AI platform, sourced to an official court page, and the 2022 opinions, sourced to the SPC English page. The platform quote is “digital assistant of judges”; the 2022 policy quote is “rulings must always be made by judges”. Those two claims sit side by side in the official record.

Spain’s five Ministry of Justice rows include plain-language conversion, judicial-resolution summaries, judicial-hearing transcription, DELFOS search and an AI-use policy. Estonia’s five RIK rows include court-ruling anonymisation, court-document labelling, hearing transcription, prison e-services face recognition and e-Notary identity recognition.

Procurement rows are rare. Norway’s Digital Courts framework is recorded both as procurement notice and contract award, both from TED. The EU DrugDetect award records automatic drug detection in correctional institutions and prisons. The three TED amount fields stay in their currencies and appear only in Appendix A.

The one-row jurisdictions add older and narrower records. Argentina’s PROMETEA row records a public-prosecution document-automation system; Malaysia’s AiCOS row records sentence recommendations in Sabah and Sarawak; Germany’s court-file structuring assistant is a contract-award row; Hong Kong’s guidelines are a judiciary policy row.

Court-disclosure and policy rows are a small but distinct category

The register has four judgement-disclosure rows. They are not policies and they are not party misconduct records. They are cases in which the court record states the court’s own AI use or a reviewing court records a judge’s AI use. The register names judges only where the opinion itself does.

DateRecordCourtJudge named in recordStatus
2024-05-28LLM ordinary-meaning research in Snell v. United Specialty Insurance Co.U.S. Court of Appeals for the Eleventh CircuitNEWSOMdisclosed in concurrence
2024-08-02ChatGPT 3.5 use in tutela decisionJuzgado del Circuito de Cartagenahizo uso de IA
2024-09-05LLM ordinary-meaning research in United States v. DeleonU.S. Court of Appeals for the Eleventh CircuitNEWSOMdisclosed in concurrence
2025-02-20ChatGPT references and judicial-use discussion in Ross v. United StatesDistrict of Columbia Court of AppealsSHANKER; HOWARD; DEAHLdisclosed in opinion

The Snell row, linked to the Eleventh Circuit PDF, records a concurrence that asks what ChatGPT thinks about ordinary meaning. The Deleon row, linked to the Eleventh Circuit PDF, records LLM research in another concurrence. The Ross row, linked to the D.C. Court of Appeals PDF, records ChatGPT use discussed by the majority and dissent.

Colombia’s T-323/24 row, linked to the Constitutional Court judgement, records that a Cartagena tutela judge “hizo uso de IA para emitir su decisión, en específico, de ChatGPT 3.5”. That row is coded as generative AI and a judgement disclosure. It is separate from Colombia’s Copilot transcription and act-drafting row, which is an implemented inventory entry by the judicial branch digital unit.

The policy and conduct rows are separate again. There are 25 policy rows and 2 judicial-conduct opinion rows. The US portion includes federal Administrative Office guidance, 18 state or local policy rows and two conduct opinions. The verification caveat matters here: Arizona’s 30 October 2024 generative-AI order, Kansas, Massachusetts, Nevada, New York, Pennsylvania, Virginia and Washington have official links recorded but are not quoted as read primary text in this report.

US state or federal bodyDateRecordVerification
Federal judiciary2025Interim guidance for Judiciary use of AI toolsRead from the official record
Arizona2024-01-24Administrative Order No. 2024-33 establishing the AI and Courts steering committeeRead from the official record
Arizona2024-03-27Administrative Order No. 2024-011 establishing an Artificial Intelligence CommitteeRead from the official record
Arizona2024-10-30Administrative Order No. 2024-207 on use of generative AI in the judicial branchOfficial link recorded, text not quoted
California2025-09-01Rule 10.430 generative artificial intelligence use policiesRead from the official record
Connecticut2024-02-01JBAPPM Policy 1013 Artificial Intelligence Responsible Use FrameworkRead from the official record
Connecticut2026-08-10Committee on Artificial Intelligence in the Connecticut Legal System final reportRead from the official record
Delaware2024-10-21Interim Policy on the Use of GenAI by Judicial Officers and Court PersonnelRead from the official record
Florida2026-01-15Administrative Order No. 26-04 disclosure of generative AI in court filingsRead from the official record
Illinois2025-01-01Illinois Supreme Court Policy on Artificial IntelligenceRead from the official record
Illinois2025-01-01Illinois Supreme Court AI Policy Judicial Reference SheetRead from the official record
Kansas2025-02-27Administrative Order 2025-CM-017 establishing an artificial intelligence committeeOfficial link recorded, text not quoted
Louisiana2025-10-03Generative Artificial Intelligence Guidelines for the judiciaryRead from the official record
Massachusetts2025-11-12Interim guidelines for use of Generative AI by judges and court personnelOfficial link recorded, text not quoted
Michigan2023-10-27Ethics Opinion JI-155 on judicial technology competence and AIRead from the official record
Nevada2025AI Guide for Judicial OfficersOfficial link recorded, text not quoted
New York2025-10Interim Policy on the Use of Artificial IntelligenceOfficial link recorded, text not quoted
Pennsylvania2025-09-09Interim Policy on the Use of Generative Artificial Intelligence by Judicial Officers and Court PersonnelOfficial link recorded, text not quoted
Virginia2025-05-07Supreme Court of Virginia Rules for Generative AI UseOfficial link recorded, text not quoted
Washington2025Board for Judicial Administration AI Statement of PrinciplesOfficial link recorded, text not quoted
West Virginia2023Judges’ use of artificial intelligence — JIC Advisory Opinion 2023-22Read from the official record

What to watch

The next changes to this register should come from inventories, not only rules. The DOJ inventory is annual; a later DOJ release would show which “Pre-deployment” and “Pilot” rows moved, stayed or disappeared. The UK Algorithmic Transparency Recording Standard is updated record by record, so OPG’s pre-deployment assistant, CPS’s beta pilot and nDelius public beta are the rows to re-check first.

Brazil’s CNJ survey status words are another watch list. Thirty-eight CNJ survey rows say “Finalizado e em produção / uso”, but 25 say “Em desenvolvimento”, 14 say “Em teste” and 4 say “Finalizado, aguardando autorização para entrar em produção / uso”. Any future CNJ publication can be matched row by row against those phrases.

Procurement records need a different test. Norway’s Digital Courts framework and the EU DrugDetect award are purchase records; the next public record may be an award notice, a call-off, an implementation note or nothing. They should not be merged with live-use rows without a later source.

The judgement-disclosure category is small. If more courts disclose their own use of LLMs in opinions, those rows should remain separate from policies and from misconduct incidents. The useful question is not whether a court mentioned AI. It is whether the court used it, what for, and how the opinion says that use was checked.

Appendix A — data tables

Country table

CountryCodeRowsGenerative AIInventory / register / survey rows
United StatesUS412116376
BrazilBR985692
NetherlandsNL31031
United KingdomGB2159
EstoniaEE500
IndiaIN500
SpainES520
United Arab EmiratesAE404
ChinaCN310
SingaporeSG210
NorwayNO200
ColombiaCO221
GermanyDE110
Hong KongHK110
European UnionEU100
MalaysiaMY100
ArgentinaAR100

Body type table

Body typeRows
Police and forensics177
Tribunal or adjudicator79
Court69
Prosecution service69
Other justice agency58
Corrections and probation56
Justice ministry41
Judiciary22
Court administration21
Legal aid body3

Record type table

Record typeRows
Use-case inventory entry412
Official survey entry92
Official announcement44
Internal policy25
Transparency record9
Court discloses its own AI use4
Contract award3
Pilot report3
Judicial-conduct opinion2
Procurement notice1

AI kind table

AI kindRows
Generative AI185
Analytics80
Search and retrieval77
Document automation69
Risk assessment40
NLP / classification35
Not specified32
Transcription27
Chatbot19
Translation14
Other11
Record linkage6

DOJ component table

DOJ componentRowsGenerative AI
Drug Enforcement Administration5518
Federal Bureau of Investigation4811
Bureau of Alcohol, Tobacco, Firearms and Explosives393
Federal Bureau of Prisons386
Department-wide203
United States Marshals Service187
Civil Division173
Tax Division144
Antitrust Division107
Justice Management Division / Organized Crime Drug Enforcement Task Forces104
Office of the Inspector General93
Justice Management Division85
Executive Office for Immigration Review84
Office of Justice Programs71
Executive Office for United States Attorneys76
Civil Rights Division71
Office of the Pardon Attorney62
Environment and Natural Resources Division50
Office of Public Affairs51
Criminal Division50
United States National Central Bureau41
United States Trustee Program22
Professional Responsibility Advisory Office20
Office of Professional Responsibility20
Organized Crime Drug Enforcement Task Forces20
Office of Community Oriented Policing Services10
National Security Division10
United States Parole Commission10

Source status words with more than one row

The complete script output records 97 distinct source status strings. The table below gives the 20 status strings that appear more than once; the 77 single-use strings are in the run log.

Status as statedRows
Deployed155
Pre-deployment103
Operation and Maintenance49
Finalizado e em produção / uso38
In gebruik31
Retired30
Pilot25
Em desenvolvimento25
Em teste14
Acquisition and/or Development7
production6
Initiated6
Finalizado, mas não está em produção6
Concepção (ideação e planejamento)5
currently used4
Service Description4
Finalizado, aguardando autorização para entrar em produção / uso4
Interim Policy2
can-standard2
disclosed in concurrence2

Policy and conduct-opinion table

Country / stateDateRecordTypeVerification
Brazil2025-03-11Resolução CNJ n. 615/2025 AI governanceInternal policyRead from the official record
China2022-12-12Opinions on AI application in judicial fieldsInternal policyRead from the official record
Spain2024Policy on the use of AI in the Administration of JusticeInternal policyRead from the official record
United Kingdom2025-11-13Internal policy for AI use by prosecution staffInternal policyRead from the official record
Hong Kong2024-07Guidelines on the Use of Generative Artificial IntelligenceInternal policyRead from the official record
India2022-11-26eCourts Phase III intelligent schedulingInternal policyRead from the official record
United States2025Interim guidance for Judiciary use of AI toolsInternal policyRead from the official record
United States — Arizona2024-01-24Administrative Order No. 2024-33 establishing the AI and Courts steering committeeInternal policyRead from the official record
United States — Arizona2024-03-27Administrative Order No. 2024-011 establishing an Artificial Intelligence CommitteeInternal policyRead from the official record
United States — Arizona2024-10-30Administrative Order No. 2024-207 on use of generative AI in the judicial branchInternal policyOfficial link recorded, text not quoted
United States — California2025-09-01Rule 10.430 generative artificial intelligence use policiesInternal policyRead from the official record
United States — Connecticut2024-02-01JBAPPM Policy 1013 Artificial Intelligence Responsible Use FrameworkInternal policyRead from the official record
United States — Connecticut2026-08-10Committee on Artificial Intelligence in the Connecticut Legal System final reportInternal policyRead from the official record
United States — Delaware2024-10-21Interim Policy on the Use of GenAI by Judicial Officers and Court PersonnelInternal policyRead from the official record
United States — Florida2026-01-15Administrative Order No. 26-04 disclosure of generative AI in court filingsInternal policyRead from the official record
United States — Illinois2025-01-01Illinois Supreme Court Policy on Artificial IntelligenceInternal policyRead from the official record
United States — Illinois2025-01-01Illinois Supreme Court AI Policy Judicial Reference SheetInternal policyRead from the official record
United States — Kansas2025-02-27Administrative Order 2025-CM-017 establishing an artificial intelligence committeeInternal policyOfficial link recorded, text not quoted
United States — Louisiana2025-10-03Generative Artificial Intelligence Guidelines for the judiciaryInternal policyRead from the official record
United States — Massachusetts2025-11-12Interim guidelines for use of Generative AI by judges and court personnelInternal policyOfficial link recorded, text not quoted
United States — Michigan2023-10-27Ethics Opinion JI-155 on judicial technology competence and AIJudicial-conduct opinionRead from the official record
United States — Nevada2025AI Guide for Judicial OfficersInternal policyOfficial link recorded, text not quoted
United States — New York2025-10Interim Policy on the Use of Artificial IntelligenceInternal policyOfficial link recorded, text not quoted
United States — Pennsylvania2025-09-09Interim Policy on the Use of Generative Artificial Intelligence by Judicial Officers and Court PersonnelInternal policyOfficial link recorded, text not quoted
United States — Virginia2025-05-07Supreme Court of Virginia Rules for Generative AI UseInternal policyOfficial link recorded, text not quoted
United States — Washington2025Board for Judicial Administration AI Statement of PrinciplesInternal policyOfficial link recorded, text not quoted
United States — West Virginia2023Judges’ use of artificial intelligence — JIC Advisory Opinion 2023-22Judicial-conduct opinionRead from the official record

Judgement-disclosure table

DateCountry / stateRecordJudge named
2024-05-28United StatesLLM ordinary-meaning research in Snell v. United Specialty Insurance Co.NEWSOM
2024-08-02ColombiaChatGPT 3.5 use in tutela decision
2024-09-05United StatesLLM ordinary-meaning research in United States v. DeleonNEWSOM
2025-02-20United States — District of ColumbiaChatGPT references and judicial-use discussion in Ross v. United StatesSHANKER; HOWARD; DEAHL

Published money figure rows

RecordCountryRecord typeAmount as published on TED when register builtCurrencyDate
IT development services framework for Digital CourtsNorwayContract award640,000,000NOK2025-09-03
IT development services framework for Digital CourtsNorwayProcurement notice640,000,000NOK2025-03-27
Automatic drug detection in correctional institutions and prisonsEuropean UnionContract award2,480,000EUR2024-02-29

Sources

Primary documents and official records cited above:

Reuse this research

Quote it, cite it, forward it — CC BY 4.0 for the data and figures; the linked official documents are the record. Suggested citation and a link to this exact report: