# TNA application preparation — Casus English

Prepared 4 October 2026. Working draft only; not submitted. No legal entity, governance arrangement, licence approval or implemented safeguard should be inferred from this document.

The proposed initial scope is a subscription research service for legal professionals studying English commercial law. If the actual launch includes direct-to-consumer advice or litigation predictions, revise the description and risk assessment openly; do not retain the narrower description merely to obtain a licence.

Application: https://caselaw.nationalarchives.gov.uk/apply-for-a-licence
Preparation guidance: https://caselaw.nationalarchives.gov.uk/what-you-need-to-apply-for-a-licence
Licensing contact: caselawlicence@nationalarchives.gov.uk

## Purpose and activities — proposed text

We propose a subscription legal research service for legal professionals working with English commercial law. The service will retrieve court judgments and relevant paragraphs, identify citations and procedural history, and provide source-linked research summaries and clearly labelled translations in the user's requested language. We seek permission to ingest and update Find Case Law records, perform text extraction, indexing, vector embedding, retrieval and citation analysis, and deliver relevant material through authenticated web, API and Model Context Protocol interfaces to approved AI assistants. Original English text, paragraph references, source links and coverage limitations will accompany research outputs. We propose human review of treatment classifications and high-impact summaries, with correction and withdrawal procedures. The service is intended to support professional research, not provide autonomous consumer legal advice, litigation outcome predictions or profiles of individuals. Details of external processors, retention and permitted downstream use will be agreed before operation.

## Public statement — proposed text

Casus English is a proposed multilingual research service for professionals working with the law of England and Wales. It aims to make published court judgments easier to find, compare and verify within the AI tools professionals already use. Users will be able to ask questions in their preferred language and inspect supporting English passages and source references. The proposed methods include lexical and semantic search, citation extraction and carefully evaluated AI-assisted summaries and translations. The service will describe its coverage and limitations, distinguish original judgments from generated content, and preserve human professional responsibility for legal conclusions. Its proposed safeguards include access controls, data minimisation, regular quality evaluation and procedures for updating or removing revised or withdrawn judgments.

## Facts to supply before submission

- Full legal name, trading names, registration number, registered country and address.
- Senior person responsible for licence compliance and contact details.
- Proposed product URL, or an accurate statement that none is publicly available.
- Customer groups, permitted uses, access model and launch jurisdictions.
- Actual infrastructure locations, processors and AI providers; distinguish embedding, reranking, summarisation, translation and host-assistant providers.
- Data flow diagram, including material returned to an external assistant and retained in user conversations.
- Actual, rather than aspirational, security, privacy, retention and incident-response arrangements.
- Names and qualifications of reviewers and any independent oversight; leave these unresolved if nobody has yet been engaged.
- Honest answers about publication of full judgments, extracted data and methodology.

The two statements above describe proposals. Convert a safeguard to a claim about an implemented control only after it exists and has evidence.

## Proposed measures against the nine principles

These are implementation proposals requiring review and evidence, not completed compliance certifications.

| Principle | Proposed control | Evidence needed |
|---|---|---|
| Dignity of the court | Display original context, distinguish commentary and translation, no misleading official branding | Sample outputs and editorial policy |
| Independence of the court | No personal judicial rankings or predictions based on an individual judge | Product scope, tool restrictions, misuse tests |
| Appropriate scrutiny | Research focuses on law and reasoning; avoid constructing profiles of parties or lawyers | Data model, use-case assessment and documented exceptions |
| Anti-discriminatory harm | Exclude background-check and criminal-record screening products; investigate discriminatory outputs | Abuse policy, evaluation cases and escalation process |
| Anti-bias | Evaluate topic, court, temporal and language coverage; regularly review model changes and retrieval failure | Versioned benchmark, sampling plan and reviewer records |
| Personal privacy | Assess lawful basis and sensitive data; minimise processor inputs; preserve anonymity orders and apply takedowns | DPIA assessment, data-flow map, contracts and deletion test |
| Discoverability | Authenticate document access; restrict automated extraction; no search-engine indexing of licensed judgment bodies | Access-control tests, rate limits, crawler controls |
| Algorithmic transparency | Explain retrieval, generated content, treatment review and limits to users | Public methodology and system/version records |
| Accurate data representation | Clearly disclose incomplete coverage; attach source/paragraph/version; distinguish absence of evidence from validation | Coverage register, citation tests and correction workflow |

Do not promise complete anonymisation while retaining all party names in embeddings and external prompts. Distinguish minimum identification needed for conventional legal citation from profiling, and ask TNA to approve the concrete design. Commercial cases may still contain sensitive data.

## Questions to resolve with TNA before accepting commercial scope

1. Can the agreed Purpose expressly cover a paid professional research service using RAG, lexical indexing, embeddings, reranking, extraction of citation relationships, summaries and multilingual translations?
2. Can licensed passages be returned through authenticated MCP/API calls to specified third-party assistants, including ChatGPT, Claude, a Grok API implementation and self-hosted model clients? How should this be distinguished from prohibited third-party crawling and scraping?
3. What authorisation is required for external embedding, reranking and generation providers acting as processors? Are there restrictions on hosting regions, subprocessors, retention, logging or provider training?
4. Which end-user activities are permitted: reading, preparing advice, exporting a cited research memo, sharing that memo with a client, and retaining output in an assistant conversation?
5. Does any part of the proposed downstream delivery require written sublicensing consent or separate licences for enterprise customers or platform partners?
6. What counts as an acceptable method of withdrawing revised or removed judgments from caches, vectors, generated summaries, translations, backups and user-accessible exports? What is expected for copies already retained in independent third-party chats?
7. How should citation identifiers and party names necessary for legal research be treated under anonymisation expectations? What controls are expected before transmitting passages to external models?
8. May we publish paragraph-linked original text behind login, reviewed treatment evidence and labelled translations? Are there approval requirements for attribution and coverage statements?
9. What initial bulk delivery method and operational rate are preferred? Is there a supported change/removal feed, and how should we reconcile withdrawn records?
10. Does the standard five-year, worldwide, royalty-free transactional structure apply to the proposed operator and activities? Which obligations survive expiry, and which derived materials may be retained?
11. Would a limited experimental licence be required before prototype indexing? If R&D is granted, what additional permission is needed for a client-facing commercial pilot?
12. Is retrieval that supports a legal professional's wider expert process acceptable within the proposed commercial scope, and what safeguards distinguish it from the listed high-risk advisory uses?

A positive answer should be reflected in the executed licence or written clarification tied to it; a general marketing statement about open data is insufficient for these implementation details.

## Implementation boundaries proposed for the initial application

- Professional research with review by the user's legal professional; no automatic litigation decision-making.
- No sale of a raw downloadable TNA mirror, model-training dataset or unrestricted pass-through API.
- Controlled source excerpts and case retrieval through authenticated tools; any full-document retrieval described explicitly.
- No training or fine-tuning of general-purpose models on the corpus within the initial proposed scope. Embedding and inference remain computational analysis and must be disclosed.
- Commercial terms must not claim ownership of judicial text or grant rights the operator has not received.
- Statutes, private law reports and any later added jurisdictions keep separate provenance and rights records.
- Licence approval is distinct from personal-data compliance and from authorisation for reserved legal activities.

## Documents reviewed

- [TNA licence process](https://caselaw.nationalarchives.gov.uk/licence-application-process).
- [Open Justice Licence v2](https://caselaw.nationalarchives.gov.uk/open-justice-licence/version/2).
- [FOI publication of licence templates and MoJ guidance](https://www.nationalarchives.gov.uk/freedom-of-information/requests/cas-272267/).
- [Transactional template](https://cdn.nationalarchives.gov.uk/documents/find-case-law-transactional-licence-template.dotx).
- [MoJ guidance](https://cdn.nationalarchives.gov.uk/documents/CAS-272267-rd-transactional-licenses%20-moj-guidance-find-case-law.docx).

The downloaded template is historical public evidence, not a licence issued to this project. Current individual terms must be confirmed with TNA.