Cartoon of a hand checking a legal document with a magnifying glass beside a laptop displaying an AI assistant.

AI is already changing how law firms work, how clients communicate and how legal services are delivered. The challenge for firms is to understand where it is being used, what risks it creates and how to manage it effectively.

That was the central message of JBL Compliance’s October webinar, Beyond AI hallucinations. Hosted by Ed Marshall, the discussion brought together Jonathon Bray, Managing Director Sophie Cisler and new Head of Risk and Compliance Holly Heath Jones. Together, they explored the practical questions facing firms as AI becomes part of everyday practice.

Although invented cases and inaccurate answers remain serious concerns, the panel encouraged firms to look more broadly. Confidentiality, supervision, staff behaviour, client expectations and pricing all require attention. Having an AI policy is a starting point; knowing whether it works is the next challenge.

The audience polls reinforced that message. Most respondents reported policies and approved products, but fewer reported a requirement to identify AI-produced work. 

Established duties in a changing environment

The discussion opened with the SRA’s Misuse of AI warning notice, published on 17 August 2026. Sophie identified hallucinations, confidentiality, data protection, privilege and supervision as key themes.

Her concern was that people who understand their professional obligations may not always connect those obligations with the technology they are using. Uncertainty about how an AI tool operates can obscure questions about where client information goes and how it is handled.

The panel cautioned against assuming that a paid product, or a tool embedded within an existing office suite or case management system, is automatically suitable for confidential work. The continuing confidentiality duty appears in paragraph 6.3 of both the Code of Conduct for Solicitors and the Code of Conduct for Firms.

Jonathon framed this as principles-based regulation responding to technological change. The SRA Principles remain the foundation, particularly Principles 1, 2, 4, 5 and 7: the administration of justice, public trust, honesty, integrity and clients’ best interests.

Understand the tool before introducing it

Holly stressed the importance of preparation. Introducing AI involves more than choosing a product, writing instructions and telling staff to begin using it.

Those responsible for systems and controls need to invest in their own understanding. What does the tool do? Where could it create difficulties? What controls are needed, and how will the firm know whether those controls are effective?

This connects directly with paragraphs 2.1 and 2.5 of the Firms’ Code, covering effective governance, systems and controls, and the identification, monitoring and management of material business risks.

The final audience poll illustrated the difficulties. 29% identified keeping up with changing tools as their firm’s primary AI risk challenge, while 23% selected shadow AI. A further 17% selected a lack of practical guidance.

A policy should explain what staff can use, for which purposes and within what boundaries. Training must make those expectations understandable in everyday work. Firms then need to test whether actual practice reflects the policy.

Find out what people are already using

One of the strongest themes was the risk of “shadow AI”: tools used by staff without the firm’s knowledge or approval.

Jonathon suggested that the knowledge gap may be among the biggest current risks. If management does not know which systems people are using, it cannot properly assess the risks or establish effective controls.

This can include someone seeking a quick answer through an app on their phone, or an approved tool being used for an unauthorised purpose.

The polls showed that many respondents’ firms had established formal boundaries. 85% said their firm had an AI policy and 89% reported a list of permitted AI products. The next question is whether those boundaries are understood and followed.

Jonathon described firms offering a temporary amnesty so staff could disclose previous use of unapproved systems without fear of getting into trouble. The purpose was to establish an honest picture of existing behaviour.

He also encouraged firms to consider the incentives behind that behaviour. Unrealistic billing targets, limited access to approved tools or management’s reluctance to engage with AI may encourage staff to find their own solutions.

Make supervision open and practical

The supervision poll added a significant finding: only 40% of 55 respondents said their firm required people to identify legal output produced with AI. Another 42% said there was no such requirement, while 18% did not know.

That raises a practical question about how supervisors understand the work presented to them. Holly’s comparison with supervising a trainee was a useful starting point: someone must assess the output and take responsibility for what goes to the client.

The relevant duties include paragraphs 3.2 and 3.5 of the Solicitors’ Code, on competent service and effective supervision, and paragraph 4.4 of the Firms’ Code, requiring an effective system for supervising client matters.

Sophie emphasised openness. A supervisor needs to know which tool was used, how it contributed and what checking has already taken place.

There may also be a cost to efficiency. If checking an AI-generated document takes as long as producing it previously did, the anticipated saving may disappear.

Witness statements provided a particular example. Jonathon warned that AI may fill gaps or make an account appear more coherent, moving the drafting away from the witness’s evidence. This engages paragraphs 2.1 and 2.2 of the Solicitors’ Code, concerning misuse of evidence and influencing its substance.

Prepare for clients using AI

Clients using AI is already painfully common among the webinar audience. 86% of respondents had encountered clients using AI-generated arguments to challenge advice or bills, or bring complaints. Just over half encountered this regularly, with another 36% doing so occasionally.

An audience question about autonomous AI agents prompted discussion about establishing whether a communication reflects the client’s instructions.

Paragraph 3.1 of the Solicitors’ Code allows instructions from clients or properly authorised representatives. Where there is reason to doubt that instructions reflect the client’s wishes, the solicitor must satisfy themselves before acting. AI-mediated communications make that assessment particularly important.

Sophie also discussed clients asking a solicitor to “just check” an AI-generated document. Although the technology is new, clients have long brought their own drafts and requested a quick endorsement.

An apparently finished draft does not remove the need to understand the client’s circumstances or assess whether it achieves their aims. Explaining the scope of that review supports informed client decisions under paragraph 8.6 of the Solicitors’ Code.

Ed’s example of someone relying on AI answers produced from an incomplete account illustrated another challenge. A confident response may reinforce an existing view even when the tool has not received the full picture.

Use AI to clarify complaints

The panel discussed the pressure created by AI-generated complaints and data subject access requests. Long submissions and rapid rebuttals can increase the workload for those handling them.

Jonathon identified a constructive use for AI: helping to distil a lengthy complaint into its core grievances, organise information and prepare a chronology.

He cautioned against simply responding with another lengthy AI-generated document. The objective remains a clear response to the issues raised, alongside the duty to handle complaints promptly, fairly and without charge under paragraph 8.5 of the Solicitors’ Code.

Review billing and the wider business model

Ed posed a direct question: if AI reduces a three-hour task to one hour, what should the firm charge?

The billing poll showed that this remains unresolved for many respondents. 48% of respondents said their firm had not yet addressed the effect of increased AI drafting speed on billing. Another 30% reported a combination of approaches, while 16% selected “the hourly rate lives on”.

Jonathon’s response was that time recording must reflect the time actually spent. Firms then face a commercial decision about accommodating greater efficiency.

The panel discussed fixed fees and value-based pricing. The service’s value includes professional judgement, understanding the client’s circumstances, checking outputs and communicating advice.

Sophie challenged the assumption that AI always produces substantial time savings. Holly highlighted the technology’s cost and the importance of explaining how it fits within fees. Clear pricing information remains a requirement under paragraph 8.7 of the Solicitors’ Code.

Begin with the firm’s needs

The closing discussion encouraged firms to approach AI as a coordinated business project, connecting compliance, legal competence, finance, technology and client service.

Holly advised firms to examine their own work first. Where are the bottlenecks? Which tasks would benefit from support? What evidence is there of that need, and what risks accompany it? Those questions should guide product selection.

Staff involvement is important too. People doing the work can identify practical needs and are more likely to support changes they have helped shape.

Finally, the panel asked firms to consider what they want to preserve, for example: human relationships, empathy, individuality and the development of future lawyers. JBL Compliance’s decision to retain human delivery of training illustrated that adopting AI can involve deliberate limits.

The webinar left firms with a practical challenge: understand existing use, choose tools purposefully, train people and test what happens in practice. The polls give that challenge added urgency, particularly around visibility of AI-assisted work, client communication and billing.

Want access to the recording?

Please email info@jblcompliance.com for a copy (free for 30 days following the webinar).

To be invited to future JBL Compliance webinars, please join our newsletter at jblcompliance.com