What problem does it solve?
Courts, tribunals, prosecutors and government legal teams work through very large files. A single immigration filing can run to hundreds of pages that the parties have not organised; a prosecution may rest on long recorded interviews that prosecutors have to watch while taking notes manually; Brazil's Federal Supreme Court drafts case reports and headnotes for the appeals it decides. Finding, ordering and summarising this material is slow, manual work that comes before the legal analysis starts.
Summaries are an obvious help and an obvious risk. A summary that omits a key fact, misattributes a statement or invents a citation can distort a decision about someone's liberty, residence or rights, and the High Court of England and Wales has already dealt with fictitious case citations, suspected to come from generative AI, put before it. The design has to keep every summary traceable to the record and every judgment with a person.
How does it work?
- Ingest the file. Filings, bundles, evidence recordings and earlier decisions are uploaded in the secure environment; scans are converted to text and recordings are transcribed with time stamps.
- Organise. Documents are classified by type (application, evidence, submission, decision), tabbed and deduplicated, so the record has a navigable structure.
- Summarise with references. The model produces a summary in an agreed template (parties, key facts, chronology, issues, relief sought) where every statement points to the page or time stamp it came from.
- Draft routine documents. For high volume work it drafts standard documents such as case reports or headnotes for a staff member to review and adapt.
- Verify and decide. The judge, prosecutor or lawyer checks the summary against the record and makes every decision; feedback on errors improves the templates.
- Audience
- Employee facing
- Autonomy
- Copilot
- Adoption
- Early adopters
- Channels
- Internal tools
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
No public deployment has disclosed a measurable outcome yet.
Value drivers: Employee productivity, Speed and cycle time.
Indicative value
A tribunal or prosecution service that reviews 20,000 case files a year
EUR 250,000 to EUR 4 million
Legal staff time released per year
How this is calculated
Formula: files * hoursSaved * hourlyCost. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Case files reviewed per year files, case files per year | 10,000 | 30,000 | Editorial assumption. For scale, the Crown Prosecution Service expects to use its video evidence tool on about 11,000 cases a year. Source |
| Staff hours saved per file after checking hoursSaved, hours per file | 0.5 | 1.5 | Editorial assumption. No organization on this page publishes a measured saving; the person must still read the underlying material. |
| Fully loaded cost of a lawyer's or case officer's hour hourlyCost, EUR per hour | 50 | 90 | Editorial assumption. Replace with your own staff cost. |
What it leaves out: Time released, not cash saved. It leaves out the value of shorter backlogs and faster decisions for the people involved, the cost of secure infrastructure and assurance, and the cost of an error that reaches a decision.
Who already uses it?
4 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
Crown Prosecution Service
United Kingdom · Government and public sector · 2026
Prosecutors in England and Wales review recorded video interviews as part of case preparation. Beam Notes lets them upload a recording and receive a time stamped transcript and a structured summary (key details such as names and dates, and a chronology of events) built on templates that the CPS co designed with the supplier. Prosecutors must still watch the original video, every summary is reviewed before anything is moved into case systems, users rate summaries in the app, and the tool gives no advice on charging or case outcomes. The transparency record says it will be used across all 14 CPS areas on about 11,000 cases a year, with data kept by the supplier for up to 30 days.
No outcome disclosed.
Gemeente Amsterdam
Netherlands · Government and public sector · 2026
The City of Amsterdam's legal department keeps an internal case library of its advice on objections (bezwaren) against municipal decisions. The city has registered a tool, built in house, that would use an OpenAI GPT model (gpt-3.5-turbo or gpt-4) through Azure to write a summary of the core of each existing advice, shown first in the library, so lawyers handling a new objection can judge more quickly whether an earlier advice is relevant. Users can report errors in a summary, which are then corrected; the register says summaries will be marked as made with generative AI and checked by sampling. The register entry is marked "In gebruik" (in use) with a start date of July 2022, but its method section says the algorithm still has to be developed and that the prompting approach is yet to be decided, so the tool is recorded here as announced. No outcome figures are published.
No outcome disclosed.
Supremo Tribunal Federal
Brazil · Government and public sector · 2025
Brazil's Federal Supreme Court runs Maria, a generative AI platform inside its STF Digital environment that supports court staff in analysing cases and producing documents. It started by generating headnotes (ementas) in the National Council of Justice's standard format, case reports (relatórios) for extraordinary appeals and questionnaires for initial petitions in constitutional complaints (reclamações), and has been extended to more case classes, grammatical review and a unified search of related precedents. The court stresses that Maria never acts autonomously: staff review, adapt and decide whether to use each report or headnote. It builds on earlier tools, Victor (2018) for triaging extraordinary appeals and VitorIA (2023) for grouping similar cases. The court is deploying open source language models on servers that will soon be available in its own data center. No outcome figures are given in the article.
No outcome disclosed.
U.S. Department of Justice, Executive Office for Immigration Review
United States · Government and public sector · 2025
The Executive Office for Immigration Review, which runs the US immigration courts, lists a pre deployment use case for summarising court filings. Filings can run to hundreds of pages and are often poorly organised; the planned tool would summarise their contents with references to the source of the information, tab and label submission types, point adjudicators and legal support staff to where relevant content sits in the record, and summarise case law for training material. The stated aim is to let adjudicators spend their time on legal analysis and conclusions.
No outcome disclosed.
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Digital case files, or scans that can be converted to text reliably
- Templates for the summaries and standard documents, agreed with users
- A set of files with expert summaries to evaluate against
Systems to integrate
- Case management system of the court, tribunal or prosecution service
- Document and evidence stores, including audio and video evidence
- Identity and access management with need to know permissions per case
Complexity: Medium
Summarisation itself is mature. The work is in secure hosting for highly sensitive material, templates agreed with the judiciary or legal profession, reliable source references, and a culture where summaries support rather than replace reading the record.
- 1
Start with staff facing preparation, not decisions
Begin with summaries that help staff navigate a file, such as the Crown Prosecution Service's summaries of video interviews, or the summaries of earlier objection advice that Amsterdam has registered for its lawyers' internal case library.
- 2
Agree templates with the users
Co design the summary structure with judges, prosecutors or lawyers, as the CPS did with its supplier, so the output fits the way they work.
- 3
Require references for every statement
Make each summary point link to the page or time stamp in the record, so checking is quick and omissions show.
- 4
Evaluate against expert summaries
Measure omissions and factual errors on a sample of files summarised by experienced staff before scaling.
- 5
Train users on limits
Mandate training before access and make clear that the summary never replaces reading or watching the evidence, as the CPS requires.
Guardrails
- The person responsible reads or watches the underlying material; the summary is an aid
- Every summary statement is referenced to the record
- No generation of legal authorities; case law is retrieved from trusted databases and verified
- Hosting and retention appropriate to the sensitivity of the case material
- AI generated content labelled as such in the case file
KPIs to instrument
- Time to prepare a case for review, before and after
- Omission and error rate on a sampled set of summaries
- Share of summaries rated usable without major edits
- Backlog and time to decision
- Errors reported by users, and time to fix templates
Human in the loop
Judges, prosecutors and lawyers review every summary against the record and make every decision. Summaries are never placed in the case record or shared with parties without review, and users can flag and correct errors.
Common failure modes
- Omitted facts
- A summary that leaves out a key fact or a contradiction in the evidence can steer the reader. Require references and sample for omissions.
- Invented authorities
- Generative models can produce plausible but fictitious case law. Only cite from trusted legal databases and verify every authority.
- Over reliance
- Under time pressure staff read the summary instead of the record. Make reading the source part of the process and audit it.
- Transcription errors in names and details
- Speech recognition misspells names or mishears details in recorded evidence. Let users correct transcripts and check key details against the recording.
What are the risks and rules?
EU AI Act
Depends on design
Annex III point 8(a) makes AI high risk when it is intended to assist a judicial authority in researching and interpreting facts and the law and in applying the law to a concrete set of facts. Tools for prosecutors fall under point 6(c) if they evaluate the reliability of evidence, and tools that assist the examination of asylum, visa or residence applications fall under point 7(c). Under Article 6(3) a system that only performs a narrow procedural task or a preparatory task, such as organising a file or transcribing and summarising it for the person who decides, may not be high risk, but the provider must document that assessment (Article 6(4)). Summaries of internal legal advice for government lawyers, as Amsterdam plans, are generally outside Annex III.
Rules that apply
Guidance
- Artificial Intelligence (AI) judicial guidance (October 2025) (Courts and Tribunals Judiciary of England and Wales, Europe). Guidance for judicial office holders on confidentiality, hallucinations and bias; it lists summarising large bodies of text as a potentially useful task, provided the summary is checked for accuracy.
- EU AI Act Annex III, high risk AI systems (European Union, Europe). Point 8(a) covers AI used by or for judicial authorities to research and interpret facts and law; points 6(c) and 7(c) cover evaluating evidence in criminal cases and examining asylum, visa and residence applications.
Controls to put in place
- Documented Article 6(3) assessment or high risk conformity work for court facing tools
- Transparency record and user guidance for every tool
- Mandatory user training before access
- Sampled quality review of summaries against the record
- Retention and access controls matched to case sensitivity
When it went wrong elsewhere
- High Court of England and Wales: Ayinde v London Borough of Haringey and Al-Haroun v Qatar National Bank [2025] EWHC 1383 (Admin). The Divisional Court dealt with two cases in which fictitious or inaccurate case citations, suspected to come from generative AI, were put before the court, and warned the legal profession about its duty to verify authorities.
Frequently asked questions
- Do courts use AI to summarise case files?
- Some do, with staff in control. Brazil's Federal Supreme Court uses its Maria platform to draft case reports and headnotes for staff to review, the UK Crown Prosecution Service summarises video interviews with Beam Notes, and the US immigration courts have listed filing summaries as a planned use.
- Is AI summarisation for judges high risk under the EU AI Act?
- It can be. Annex III point 8(a) covers AI that assists judicial authorities in researching and interpreting facts and the law. A purely preparatory tool, such as one that organises and summarises a file, may fall under the Article 6(3) exception, but the provider has to document that assessment.
- What is the biggest risk?
- That a summary or a generated citation is trusted without checking. English courts have already dealt with fictitious authorities put before them. Keep summaries referenced to the record and verify every authority.
How to cite this page
Blits.ai AI Use Case Library, "AI for court and case file summarization", last verified 27 September 2026, https://www.blits.ai/ai-use-cases/court-and-case-file-summarization. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 27 September 2026: First published