Anna Batallas and Chloe Helps consider the SRA warning notice on the use of AI issued on 17 August 2026.
Imagine the following scenario: You are a trainee or junior solicitor working in your firm's commercial litigation group. You are dealing with multiple competing deadlines, and an urgent legal research task arrives in your inbox. You know your colleagues and clients are encouraging the use of AI to improve quality service delivery at lower cost, and, to that end, your firm has recently adopted a specific AI tool. You use that tool to prepare a first draft of a research note, and the output looks polished at first blush and saves you considerable time. You submit the work to your supervisor but omit to mention that you used AI to complete the assignment, and without checking the underlying sources.
What could go wrong here?
The SRA’s warning notice
Solicitors and law firms are increasingly using AI in their day-to-day work. AI tools are now commonplace in many law firms, as well as our everyday lives. The Law Society’s statistics show that eight months ago only 46% of UK lawyers used generative AI, now 61% use some form of AI. That is likely to increase. As adoption continues to grow, so too do the risks.
AI can be a valuable tool for trainees and junior solicitors, but it is important to exercise caution. The SRA has received 42 reports of AI misuse in the past year. Those concern the use of AI for harmful, unethical or unintended purposes. The SRA has identified two key concerns: hallucinations, which in simple terms are responses generated by AI that contain false or misleading information; and instances of confidential client information being entered into AI tools that lack appropriate safeguards. The reported cases of AI misuse highlight failures by solicitors to comply with their professional duties. The SRA has ‘warned’ the profession of their obligation to follow all relevant rules and regulations, particularly where AI is concerned.
Hallucinations
AI can produce ostensibly convincing work product in little time, but legal authorities and case citations may be inaccurate (at best) or entirely fabricated (at worst). Many hallucinations reflect seemingly definitive answers generated by AI where the law is ambiguous or evolving.
Concerns regarding the use of AI are particularly relevant to trainees and junior lawyers. In Cork & Anor v Smith [2026] EWHC 1199 (Ch), a junior associate was determined to have delegated the “entire thinking process” to AI, resulting in the presentation of fictional authorities to the Court. Much has been written about the case – in short, considerable risks arise from a failure to check underlying sources, apply appropriate supervision, and exercise independent judgment, which is essential to the practice of law more broadly. The consequences in Cork were severe, including self-referral to the SRA by the law firm acting for one of the litigants.
As of 10 July, there have been 67 recorded cases of AI hallucinations in the UK alone. This begs the question: how big is the overall number of cases in circumstances where it is likely that there are many (as of yet) unreported cases, and how much bigger will that number become considering it is still early days in the adoption of AI by law firms conducting litigation?
Regulators, courts and other supervising bodies are clearly treating AI hallucinations as a serious and evolving issue. It is therefore expected that legislative or regulatory frameworks will be developed to govern the use of AI in the legal profession and in other professions connected with litigation.
Confidentiality and agentic AI
Public AI tools (such as the Claude consumer AI tool) present a significant confidentiality risk. As we have explained in our piece on generative AI, enterprise (or closed) AI tools (such as your firm’s business CoPilot AI tool) provide much stronger confidentiality protection than ChatGPT, but it is still important to educate legal teams, implement clear AI usage policies, and involve lawyers early where legal issues arise.
The use of Agentic AI, being a system capable of autonomous, goal driven and adaptive behaviour with limited human input, raises further issues, and leads to the need for further safeguards: control what agents can access, ring-fence sensitive data from general productivity tools, treat prompts and outputs as part of the documentary record, and, once again, retain legal oversight over the process.
Given the SRA's concern that professionals are not always identifying and mitigating AI-related risks, organisations should implement the measures to support the safe, responsible and legally compliant use of this technology.
Implications for legal professionals
The SRA has reminded solicitors of their obligations under the Solicitors Code of Conduct and issued a warning that there will be implications in case of a breach of regulatory requirements and duties.
Key steps for trainees and junior lawyers
Before using AI for legal research, first pause and ask yourself how you intend to use the tool, and whether you understand its limitations and know your firm’s guidelines and approved processes on the use of AI. Are you clear on the risks? Have you read your firm’s AI policy, if there is one? Have you read the SRA’s warning notice and other guidance?
AI use will only increase as law firms build their own tools, Harvey and Legora expand their outreach, and others establish their own ‘AI-native’ firms. As trainees and junior lawyers, we can be at the forefront of its responsible use: prompt carefully, iterate by checking comprehensively the underlying substance and applicable law, and repeat.
1. Complete AI training and follow firm policies.
Ensure that you have (a) completed your firm’s AI training and (b) reviewed any relevant internal policies. Responsible use begins with knowing when AI can be used, how it should be used, and what safeguards are required.
2. Use AI as a starting point only.
AI can assist with first drafts, summaries, legal research, ideas and administrative tasks, but it should not replace our critical thinking or professional judgement. Treat AI-generated work as a preliminary output, rather than a final answer. It is crucial, when using AI, that we understand its capabilities and limitations.
3. Verify everything
Check every authority, quote, source and factual statement produced by AI. The tools currently in the market are particularly prone to producing inaccurate case law and their summary of the points expressed in those cases is often wrong. Read each case closely to make sure the final work product is correct. This responsibility for erroneous work falls at all levels, on trainees, solicitors, supervising partners, and the broader firm.
4. Be transparent with supervisors.
Tell your supervisor that you are using AI. In many firms that will now be assumed, but it is nevertheless important to be open and transparent. A supervising fee earner will be in a much better position to undertake effective second level review when they know that the above-mentioned steps have been followed, and it helps greatly if we have been open about our use of AI. Trust is difficult to build but easy to lose.
5. Escalate errors immediately.
If you make a mistake in your assignment after submission – and it can happen from time to time given the demands of our jobs – raise it with your supervisor immediately. This principle applies to all trainee and junior solicitor work, but it is particularly important where AI has been used given the risk of misleading others, be it a fee earner, client, or judge. Early escalation allows for early correction.
Everything in the world of legal technology is developing at lightning speed. As trainees and junior solicitors, we stand out by staying informed, using AI responsibly and ensuring that our professional obligations remain front and centre. The challenge for us is not to avoid AI - it can help us work much more effectively - but we must learn how to use AI responsibly and reading the SRA’s warning notice and your firm’s policies is a good place to start.
