AI developers are buying licensed copies of real business records to train and evaluate models, and law firms sit on years of exactly that material. Richard McKay, who introduces UK firms to those buyers, sets out why most practices should still say no.
Law firms hold years of digital work product: matter files, advice drafts, pleadings, research memos, emails and shared drives. AI companies increasingly seek licensed copies of real business records to train and evaluate models. A small number of practices may be able to do this safely, but for many firms the confidentiality and data-protection obstacles will outweigh any benefit.
What does the SRA’s AI warning notice require?
On 17 August 2026 the Solicitors Regulation Authority published its warning notice Misuse of AI. It applies to all firms and individuals the SRA regulates. Two concerns dominate: AI “hallucinations” that put false citations or inaccurate material before courts and clients; and confidential client information entered into AI systems that lack suitable contractual and technical safeguards. Our earlier report on that notice covers how the same risk arises in everyday intake, email and document review.
The notice is explicit on confidentiality. Client information should only go into AI systems where appropriate safeguards exist. Firms should satisfy themselves that client data remains in a secure environment, is not accessed by unauthorised third parties, is not used to train AI models except where explicitly authorised and appropriate, and is not retained longer than necessary. The same notice points firms to paragraph 6.3 of the Codes of Conduct.
Licensing records so an AI developer can train a model is disclosure and further use, not a closed, firm-controlled drafting tool. If training is proposed, the firm must show authorisation, including client consent where required, and safeguards that match the SRA notice.
How far does the paragraph 6.3 duty reach?
Paragraph 6.3 of the SRA Code of Conduct for Solicitors, RELs, RFLs and RSLs, and the matching provision for firms, requires you to keep the affairs of current and former clients confidential unless disclosure is required or permitted by law or the client consents.
The SRA’s confidentiality guidance underlines how wide that duty is. It is not limited to “do not email the file to a stranger”. The courts have treated it as a duty not to misuse confidential information without consent, including making use of it, or causing others to use it, otherwise than for the client’s benefit. The duty continues after the retainer ends. Consent, where relied on, must be clear: the client should understand to whom information may go, when, and for what purpose.
For an AI licence that usually means identifying what would leave the firm; deciding consent versus exclusion; recording who decided; and refusing training rights over client material without authorisation.
What does UK GDPR require after the 2025 Act?
Even where professional confidentiality can be managed, UK GDPR still applies to personal data in the archive.
The purpose-limitation principle in Article 5(1)(b) requires personal data to be collected for specified, explicit and legitimate purposes and not further processed in a manner incompatible with them. Section 71 of the Data (Use and Access) Act 2025 amended that principle and inserted Article 8A into the UK GDPR, which now sets out how to decide whether a new purpose is compatible with the original one. Section 71 has been in force since 5 February 2026.
Re-using matter data, collected to deliver legal services, so that it can be licensed to an AI developer is a new purpose. Under Article 8A the firm must take into account, among other things, any link between the original and new purposes, the context in which the data was collected including the client relationship, the nature of the data including special-category and criminal-offence data, the possible consequences for the people concerned, and safeguards such as encryption or pseudonymisation. Where the data was originally collected on the basis of consent, the routes to compatibility are narrower. The ICO has updated its guidance on compatibility and the reuse of personal information to reflect these changes.
Compatibility is not the end of the analysis. A compatible purpose does not on its own give a lawful basis for the processing. The firm still needs a lawful basis under Article 6 for licensing and disclosing the data, a condition under Article 9 for any special-category data, and must meet its transparency obligations to the people in the records.
If personal data would be shared with another organisation as a controller, the ICO’s data-sharing code also applies. Anonymous information falls outside UK GDPR, but “anonymous” is a high bar. Redaction that still leaves people identifiable is not anonymisation. Map personal data before export; prefer truly non-personal records; and do not treat a buyer’s “cleaned” claim as a substitute for the firm’s own assessment.
When should a firm decide not to license at all?
Not every archive should go to market. Common stop signs include:
- Live or recent contentious matters where disclosure could prejudice clients or waive privilege.
- No workable consent route for the clients whose affairs appear throughout the files.
- Special-category or highly sensitive personal data that cannot be removed with confidence.
- Third-party or counsel material the firm does not own and cannot license.
- Buyer terms requiring model-training rights over identifiable client content without explicit, informed client authorisation and safeguards that meet the SRA warning notice.
- No COLP or partner ownership, inventory, exclusion list or written approval of scope.
The SRA’s August 2026 news release recorded 42 reports related to potential misuse of AI between July 2025 and July 2026, with investigations including confidentiality issues. “Everyone is doing AI” is not a defence.
What should a firm do if it proceeds?
Where a firm has worked through the stop signs and still wants to go ahead, the sequence matters.
- Inventory. Record the systems and years in scope.
- Exclude. Remove counsel, third-party and client-owned material you cannot license.
- Strip or withhold. Hold back personal data until the UK GDPR position is clear.
- Get sign-off. Partner or COLP approval on purpose, buyer type and training-use wording.
- Obtain consents. Any client consents required under paragraph 6.3, before export.
- Contract. Permitted use, security, retention and no onward training use beyond what was authorised.
- Export and record. Send only the approved subset and keep a record of what left.
AI licensing of firm records can be done carefully, but for many firms the right answer will be not to license. The SRA’s 17 August 2026 notice and paragraph 6.3 set the regulatory frame. The UK GDPR, including Article 8A and the ICO’s guidance on compatibility and reuse, sets the data-protection frame. When in doubt, do not license.