AI in Legal Practice: What the SRA’s Warning Means for Law Firms

The two risks the SRA is most concerned about

The warning notice focuses on two areas.

First, inaccurate or fabricated material in legal work, especially court submissions. Generative AI can produce convincing but false case citations, quotations, summaries and legal propositions. These errors are often described as “hallucinations”, but the label should not make the consequences sound trivial.

The SRA points to recent cases in which false authorities were put before the courts. It notes that lawyers remain responsible for verifying material submitted in their name and that false citations can lead to wasted-costs applications, regulatory referrals and, in some circumstances, potential contempt of court.

Second, client confidentiality. Entering client information into an unsuitable AI tool can expose sensitive data, breach confidentiality obligations and, depending on the circumstances, risk permanently waiving legal professional privilege.

These risks apply to both free and paid-for AI systems. The relevant question is not simply whether a product is commercial or enterprise-labelled, but whether the contractual, technical and organisational safeguards are appropriate for the information being processed.

AI does not take responsibility away from lawyers

The central principle is simple: solicitors remain responsible for the advice and work delivered to clients even where AI has been used in preparing it.

AI has no separate legal personality and cannot carry professional accountability. Lawyers must apply their own informed judgement, and supervisors remain responsible for ensuring that work undertaken by junior or non-authorised colleagues receives effective oversight.

For firms, this means AI should not be treated purely as an informal personal-productivity tool when it is used in client work. It needs to sit within the firm’s established systems for quality assurance, supervision, risk management and information security.

The SRA’s approach is outcomes-focused. It does not prescribe a particular AI product, policy template or technology stack. Instead, firms need to be able to show that they have identified the risks created by their use of AI and put proportionate controls around them.

Court submissions require robust verification

The SRA says legal professionals should be assured that named case-law authorities used in submissions are genuine, relevant, verifiably cited and supportive of the arguments being advanced.

In practice, firms should treat AI-generated legal content as unverified draft material unless and until it has been checked by a suitably competent person. A sensible verification process should require the responsible lawyer to check:

• Case citations against authoritative legal sources.
• Quotations against the original judgment, legislation or source document.
• The relevance and current status of an authority.
• Whether the authority genuinely supports the proposition being advanced.
• Facts, procedural details and legal analysis before they are relied upon.

The important point is that an AI-generated error may be understandable, but it is not excused simply because it originated in a tool. Professional responsibility remains with the lawyer and, where relevant, their supervisor and firm.

Confidentiality is not solved by choosing a paid tool

A paid AI product may offer stronger contractual or technical protections than a public consumer service, but price alone does not establish that a system is suitable for confidential legal work.

Before client information is entered into an AI system, firms should understand how the service handles that information: where it is processed, who can access it, how long it is retained, whether it may be used to improve or train models, and what contractual safeguards apply. The SRA specifically highlights storage, retention, training use and unauthorised access as issues firms should understand.

A proportionate supplier and data-risk assessment could therefore examine:

• Whether client data remains within a suitably secure and controlled environment.
• Whether submitted information can be used for model training or service improvement, and under what conditions.
• Whether access is restricted to authorised users.
• Retention and deletion arrangements.
• Data-processing locations and material third-party dependencies.
• Whether contractual terms support the firm’s confidentiality, privilege and data-protection obligations.

These are practical governance controls rather than a checklist prescribed by the SRA. They are intended to help firms translate regulatory obligations into supplier, security and information-governance decisions.

The SRA’s warning is particularly clear that client information should only be entered into AI systems where appropriate contractual, technical and organisational safeguards are in place. Where a firm cannot establish what protections apply, confidential client information should not be placed into the system.

What law firms should do now

Because the SRA regulates for outcomes rather than prescribing a single implementation model, firms have flexibility in how they respond. That flexibility should not be mistaken for an absence of governance requirements.

A proportionate AI governance framework could include:

• An approved-tool register distinguishing between tools permitted for public information, internal non-sensitive work and client-confidential matters.
• Clear rules on what staff may paste, upload or prompt into AI systems.
• Documented supplier-risk assessments covering confidentiality, security, retention, training use and data protection.
• Mandatory verification requirements for AI-assisted legal research, drafting and court documents.
• Training for lawyers, support staff and supervisors on hallucinations, confidentiality and appropriate use.
• Clear escalation and incident-reporting procedures where inaccurate content or an unintended disclosure is identified.
• Regular review as products, contracts, technical controls and regulatory expectations change.

This is not a checklist prescribed by the SRA. It is a practical way of translating the regulator’s outcomes-focused expectations into controls that can be owned, evidenced and reviewed.

For in-house legal teams, there is an additional tension. The organisation may be keen to expand AI adoption for commercial or operational reasons, but that objective cannot override a solicitor’s independent professional duties. The SRA specifically advises in-house solicitors to consider whether organisational interests in expanding AI use could conflict with their own regulatory responsibilities.

A practical opportunity, if managed properly

Used well, AI can support more efficient legal services. The SRA itself recognises that firms are already using AI safely and responsibly and that the technology can support the delivery of legal services.

A blanket prohibition may still be appropriate for particular tools, data types or high-risk workflows. But as a general strategy it has limits. The SRA’s earlier Risk Outlook noted that staff may use general-purpose AI casually even where a firm has not formally adopted it. That makes visibility, clear rules and supervision important even for firms taking a restrictive position.

The sensible response is therefore neither blanket prohibition nor uncontrolled experimentation. It is controlled adoption: selecting appropriate tools, setting clear boundaries, protecting client information, maintaining meaningful human review and retaining evidence of the decisions made.

The SRA’s warning should be read as a call for law firms to treat AI governance as part of core professional practice. Firms that do so will be better placed to capture the benefits of AI while protecting their clients, their people and their reputation.