SRA Shadow AI Warning Notice: what it means for your firm in 2026

The Solicitors Regulation Authority has issued formal Warning Notices on the use of AI tools in legal practice — specifically on what the SRA calls "Shadow AI": the use of unapproved, personal or consumer-grade AI tools in professional contexts. This guide explains what the notices say, which Code of Conduct obligations are engaged, and what your firm needs to do.

1. What is Shadow AI?

Shadow AI refers to the use of AI tools that have not been formally approved or procured by the firm — typically consumer-grade tools like ChatGPT, Claude, Gemini or Copilot used through personal accounts, or business accounts that have not been vetted for data protection compliance in a legal context.

It is called "shadow" because it often happens without the firm's management or compliance officers knowing — individual fee earners using tools they find helpful without realising the professional and regulatory implications.

2. The SRA Code of Conduct obligations engaged

The SRA Warning Notices identify several specific obligations in the SRA Code of Conduct 2019 that Shadow AI puts at risk:

Paragraph 6.3 — Confidentiality

You must keep the affairs of current and former clients confidential unless disclosure is required or permitted by law or the client consents. Uploading client files to a cloud AI constitutes disclosure to a third party — the AI provider — without client consent. This is a direct breach of 6.3.

Paragraph 6.4 — Use of confidential information

You must not use confidential information for the benefit of anyone other than the client. If a cloud AI provider uses client data to improve its models, that is precisely what happens — the client's confidential information benefits the AI provider's product development.

Paragraph 3.2 — Competence

You must ensure that the work you and your firm carry out is competent and delivered in a timely manner. The SRA has indicated that using AI tools without understanding their limitations, verifying their outputs, or ensuring they are appropriate for the task may itself be a competence issue.

Paragraph 7.1 — Firm management

Managers of firms must have effective governance arrangements and clear accountability. Failing to have an AI policy that addresses Shadow AI risks — or having a policy but failing to enforce it — can engage the firm-level obligations under Paragraph 7.

3. "Hosted in the UK" is not sufficient

One of the most important points in the SRA's guidance is that a cloud provider claiming to host data in the UK does not eliminate the risks. The SRA specifically notes that:

• UK-hosted cloud services may be owned or operated by companies subject to US Cloud Act jurisdiction
• Sub-processors may be located outside the UK
• A Data Processing Agreement does not prevent government access under foreign law
• The claim "hosted in the UK" is "neither sufficient nor verifiable as a guarantee of data security"

4. What the SRA expects firms to do

The SRA's guidance sets out clear expectations for compliant AI use:

• Firms should have a written AI policy covering which tools are approved, under what conditions, and with what safeguards
• Any AI tool used with client data must be subject to a data protection impact assessment (DPIA) and a Data Processing Agreement (DPA) with the provider
• Fee earners should receive training on AI risks, including confidentiality and competence obligations
• All AI-generated output must be reviewed and verified by a qualified solicitor before being relied upon, filed or sent to a client
• Firms should consider whether client consent is required or advisable before using AI on a particular matter

5. A practical compliance checklist

6. How self-hosted AI simplifies SRA compliance

With a cloud AI tool, achieving SRA compliance requires managing a complex web of DPAs, DPIAs, sub-processor agreements and ongoing monitoring. Each new tool version or sub-processor change may require a new assessment.

With a self-hosted solution like LegisBox, the compliance picture is fundamentally simpler. Because client data never leaves the firm's network, the DPIA for client matter processing concludes with a straightforward finding: no third-party processing, no transfer risks, no sub-processor concerns. The SRA's requirements on confidentiality and data security are met by the architecture itself.