Practice Resource:
Regulatory Guidance SRA Code 6.3 • SRA Code 3.5 • LPP Retention

The SRA AI Compliance Guide for UK Law Firms

How boutique practices (5–50 staff), litigation partners, and Compliance Officers for Legal Practice (COLPs) can safely deploy frontier AI (Claude, ChatGPT, Gemini) without breaching client confidentiality or waiving Legal Professional Privilege.

Executive Answer for Fee Earners & COLPs

The SRA does not prohibit AI, but enforces strict personal liability. Under the SRA Misuse of AI Warning Notice, technological deployment does not diminish professional duties under SRA Code 6.3 (confidentiality) or SRA Code 3.5 (supervision). Commercial cloud AI tools that log prompts for 30 days expose firms to Legal Professional Privilege (LPP) waiver and US CLOUD Act subpoena warrants. Compliance requires client-side pseudonymization in local RAM before prompt transmission.

Pillar 1

Data Confidentiality & The Vendor Retention Trap (SRA Code 6.3)

Under SRA Code of Conduct Rule 6.3, solicitors are bound by an absolute regulatory duty: “You keep the affairs of current and former clients confidential unless at law or with the consent of the client you are permitted or required to do so.”

The Consumer Tier Trap

Standard consumer tiers (free or Plus accounts of ChatGPT, Claude, and Gemini) ingest prompt text directly to retrain commercial foundation models. Pasting a single client contract or correspondence constitutes an immediate data breach, an irreversible SRA Rule 6.3 breach, and an actionable violation of UK GDPR Article 5.

The Enterprise Tier Fallacy

OpenAI Enterprise and Claude Team contractually agree not to use client prompts for model training. However, standard vendor terms reserve rights to log prompts in US cloud caches for up to 30 days for human abuse inspection, exposing data to vendor employees and foreign court orders.

Contractual promises from US vendors do not relieve a solicitor of strict personal liability. If a third party can view or inspect the client’s matter details in unredacted plain text, confidentiality has been compromised.

Pillar 2

Preserving Legal Professional Privilege (LPP) under English Common Law

Legal Professional Privilege (encompassing both Legal Advice Privilege and Litigation Privilege) is an absolute, substantive right under English common law (Three Rivers DC v Bank of England [2004] UKHL 48). Privilege belongs entirely to the client, not the law firm.

The Privilege Waiver Doctrine

Under English law, communicating privileged legal advice or litigation strategy to a third party where absolute confidentiality is not maintained can operate as an implied waiver of privilege (Harman v Secretary of State for the Home Department [1983] 1 AC 280). If an adverse litigation party discovers that draft pleadings or strategy notes were uploaded to a commercial AI service where vendor personnel hold review access, they can make a formal disclosure application for the entire prompt history.

How OmitPII Retains 100% Uncompromised Privilege

Because OmitPII executes in volatile local workstation RAM, all identifying legal relationships are replaced with semantic tokens before external network dispatch:

// What leaves your workstation to the cloud AI:
"[CLAIMANT_1] seeks injunctive relief against [DEFENDANT_COMPANY] concerning breach of non-solicitation covenants under clause 14.2 of [CONTRACT_AGREEMENT_1]."

The cloud AI model performs contract interpretation and legal reasoning with full semantic fluency. However, because no client names, corporate identities, or proprietary details ever left the local machine, no privileged disclosure took place. Privilege remains unassailable.

Pillar 3

Supervision & The Anti-Hallucination Duty (SRA Code 3.5 & 2.4)

Under SRA Code 3.5, managers and partners are jointly and severally responsible for supervising the legal services delivered by fee earners. SRA Code 2.4 mandates that solicitors only make properly arguable assertions and submit factual citations to courts and tribunals.

Key Precedents Flagged by the SRA:

  • Ayinde v Haringey LBC [2024]: AI-generated fictitious citations submitted to the court resulted in severe regulatory reprimands and personal wasted costs orders.
  • Harber v Commissioners for HMRC [2023] UKFTT 1007 (TC): FTT tribunal admonished the submission of nine non-existent case summaries fabricated by ChatGPT, reinforcing that fee earners are strictly accountable for accuracy.

The Solution: Generative AI must strictly function as an assistive first-pass drafting tool. Fee earners must verify all citations against official databases (BAILII, Westlaw, LexisNexis) before any document leaves the firm.

Pillar 4

US CLOUD Act (18 U.S.C. § 2713) Immunity & UK GDPR Article 48

Under the US CLOUD Act, US federal authorities can compel US cloud providers (OpenAI, Microsoft, Google, Anthropic) to disclose data stored in their systems regardless of whether data centers reside in Dublin, London, or Frankfurt.

The Technical Inability Defense

UK GDPR Article 48 explicitly restricts data transfers pursuant to foreign warrants without a formal treaty (MLAT). OmitPII completely resolves this conflict via mathematical impossibility: because all PII redaction executes on local hardware, US cloud vendors only ever store tokenized, anonymous text. If subpoenaed, the vendor holds no identifiable matter files to surrender.

Action Plan

The 7-Point SRA AI Compliance Checklist for Law Firms (5–50 Staff)

This operational checklist provides COLPs, managing partners, and IT leaders with an immediate compliance roadmap:

1
Prohibit Consumer AI Accounts on Workstation Endpoints Block free and personal tiers of ChatGPT and Claude via DNS or browser management policy to prevent unauthorized model retraining breaches.
2
Enforce On-Device RAM Redaction for Outbound Prompts Deploy client-side proxies (OmitPII) across all Windows workstations to ensure personal names, sort codes, matter numbers, and figures are stripped before cloud transit.
3
Maintain a Mandatory Fee-Earner Verification Policy Mandate that no AI-assisted pleading, letter of claim, or contract clause is served without secondary verification by a qualified fee earner (satisfying SRA Code 3.5 & 2.4).
4
Complete a Formal AI DPIA (UK GDPR Article 35) Document systemic risks, cross-border transfer assessments, and technical mitigation controls using our pre-configured UK AI DPIA Template.
5
Audit Subprocessor Registers & Vendor Terms Verify whether cloud vendors utilize secondary subcontractors or host data outside the UK/EEA perimeter without adequate safeguards.
6
Establish Tamper-Evident Local Audit Logs Ensure cryptographic audit logs are preserved locally on firm infrastructure to demonstrate redaction compliance during SRA spot checks.
7
Deliver Annual AI Literacy Training (EU AI Act Art. 4) Train fee earners on prompt engineering best practices, hallucination detection, and prompt injection vulnerabilities.
14-Day Pilot Sandbox Windows MSI / Intune Ready

Deploy Frontier AI Safely Across Your Firm

Test OmitPII on 1 to 5 fee earner workstations in a self-contained sandbox. Zero credit cards, zero cloud connections, and zero changes to your existing network infrastructure.

• Windows 10/11 Enterprise • Sub-12ms RAM Masking • SRA Code 6.3 Aligned