AI for barristers: what chambers now have to decide

Almost everything written about AI for barristers addresses the individual practitioner. Verify your authorities. Protect client confidentiality. Do not mislead the court. That ground is well covered, including in our own analysis of what the BSB's AI guidance means for barristers, and the principles are not seriously in dispute.

The question chambers now face is different and considerably harder. Adoption across the Bar is already widespread. A good deal of it is happening outside any formal policy. And the regulator has made clear that it expects sets to hold a position rather than leaving the matter to the judgement of individual members. That is a governance and procurement problem, not an ethics one, and the chambers model makes it unusually awkward to solve.

This article sets out what the evidence actually shows about adoption at the Bar, why chambers find technology decisions structurally difficult, what the regulator expects of a set as opposed to a barrister, and what a workable chambers position needs to contain.

How many barristers are using AI?

The adoption figure deserves attention because of how quickly it has moved. The picture painted by research only a year or so earlier was of a profession approaching the technology cautiously, and the direction of travel since then has been sharp rather than gradual.

On the current figure, close to half the Bar is using these tools in some form. Whether chambers know which tools, used for what, and under what controls is an entirely separate question, and it is the one this article is about.

What the regulator's own research found

The BSB published a report, Technology and Innovation at the Bar, in 2025, carried out by Jemma Macfadyen and David Bish of Spinnaker Research and Consulting. Its findings remain the most useful account of how the Bar actually behaves around technology, and they complicate the lazy assumption that the profession is simply behind.

The research found barristers to be pragmatic rather than resistant, preferring proven solutions to novel ones, and highly aware of risk. Participants described using AI and task-specific tools for routine administrative work while emphasising that human judgement, skill and experience remain central to what they do.

On investment, chambers compared favourably with other legal services providers. Thirty-five per cent of chambers had invested in new or improved services in the preceding three years, against eighteen per cent of solicitors' firms. Chambers also showed higher uptake of certain technologies, including video conferencing and cloud-based storage, which makes sense for practitioners who may be working from home, in court or in chambers on any given day.

Significantly, the research found that participants did not regard the legal regulatory framework as an impediment to technology adoption. Whatever is slowing the Bar down, it is not the Handbook.

The problem of shadow AI

The gap between individual adoption and collective oversight has a name. Research by the Institute for the Future of Work, which has partnered with the Bar Council on work into the experiences of legal professionals, identified widespread concern about what it calls shadow AI, meaning tools used outside formal governance arrangements.

For most organisations that is an information security concern. For a profession whose entire value rests on confidentiality, accountability and compliance, it is something more serious. Use that nobody has sanctioned, nobody has assessed and nobody can audit is precisely the risk that guidance aimed at individual conduct does not reach, because the individual using the tool may be complying with every duty they are conscious of while the set has no idea the tool is in use.

The same research noted that specialist legal expertise is increasingly combined with new capabilities, including data literacy and the judgement required to interpret and validate AI outputs. That is a training question as much as a policy one.

Why technology procurement is harder at the Bar

There is a structural reason chambers lag behind firms on technology strategy, and it is not cultural conservatism. The BSB's research identified the mechanism precisely, and understanding it changes what a realistic solution looks like.

In a solicitors' firm, technology is adopted at firm level. Someone owns the decision, the budget sits in one place, and the benefit accrues to the entity that paid. At the Bar, responsibility is split between the individual and the set, and neither side of that split works cleanly.

An individual barrister adopting a tool is acting not only as counsel but as their own chief technology officer, risk manager, procurement lawyer and IT support. For a self-employed practitioner with a full practice, that is a substantial barrier, and it is a barrier that has nothing to do with willingness.

At chambers level the economics work against strategic procurement in a specific and underappreciated way: costs are shared but benefits are not. A set that licenses a research platform spreads the cost across its membership, while the gain accrues unevenly to whoever happens to use it most. In the self-employed chambers model, where members are independent of one another rather than partners in a common enterprise, that asymmetry makes collective investment genuinely difficult to agree.

What the BSB guidance expects of chambers

The Guidance on the Use of Artificial Intelligence and Other Technologies, valid from 18 May 2026, is supportive in tone rather than restrictive. The regulator states that it supports the adoption and use of technologies, including AI, where they are likely to benefit practice management, or the availability, quality or client experience of barristers' services.

What is new is where it places responsibility. The guidance addresses competence and practice management responsibilities relating to adoption, not only conduct at the point of use. It advises barristers and chambers to prepare themselves appropriately and to take a risk-based approach before procuring, adopting and using AI technologies. And it addresses use by others, meaning that clients, solicitors and opposing parties may themselves be using these tools in ways that affect a barrister's own responsibilities.

Speaking at a webinar on the guidance in May 2026, attended by representatives of 95 sets of chambers, Ewen MacLeod, the BSB's Director of Strategy, Policy and Insights, framed the chambers dimension as understanding how members are adopting new technology and what risks that adoption might generate for the set collectively.

It was, as the session noted, the first time the Bar had received explicit and specific guidance from its regulator on managing this technology.

Three layers of responsibility

It helps to separate what the guidance covers into three distinct layers, because chambers tend to think only about the second.

Before use. Preparation, competence, risk assessment, and the procurement decision itself. This is the layer that is new, and the layer that is collective.

At the point of use. The familiar duties: accuracy, verification, confidentiality, not misleading the court. This rests on the individual barrister and is not transferable.

Use by others. What happens when the solicitor instructing you, the client providing a statement, or the opposing party has used AI. This is the least discussed and arguably the fastest growing, and it affects how material arriving into chambers should be treated.

The Bar Council's position

The Bar Council welcomed the BSB guidance on 19 May 2026, noting that it complements the Bar Council's own guidance, which was updated in November 2025. The two bodies had engaged through a joint technology and innovation working group.

Kirsty Brimelow KC, Chair of the Bar Council, said the guidance would help barristers understand what the regulatory framework means for their use of AI, and encouraged conscientious and responsible use of modern technologies to help provide access to justice and assist with the administration of justice.

The distinction between the two sets of guidance matters for compliance purposes. BSB guidance sits within the regulatory framework. Bar Council guidance is practical assistance from the representative body, and it does not carry the same status under the BSB Handbook. Both are worth reading. Only one is the regulator speaking.

The enforcement picture

Two figures from the regulator show where pressure is building, and they point in different directions.

Complaints drafted with AI

The first concerns complaints made using AI rather than conduct involving it. In the introduction to the BSB's 2026 to 2027 business plan, published on 8 April 2026, Interim Director General Steven Haines wrote that the use of AI is now a contributing factor in driving reports, and acknowledged that the regulator's efficiency and timeliness targets had not been fully met as a result. Complaints to the BSB rose by approximately 25% over the year.

The practical implication for chambers is specific. Complaints drafted with AI assistance tend to be better structured and harder to dispose of on procedural grounds than traditional submissions. That does not make them more likely to succeed, but it does mean they take longer to answer and require more careful handling. A set whose complaints procedure is informal or undocumented is the most exposed to that shift.

Conduct referrals

The second concerns AI use by barristers themselves. As at the May 2026 webinar, the BSB had received around twenty referrals about barristers potentially misusing AI, then under investigation.

Context matters here and the distinctions are the ones our editorial standards require. A referral is not a finding of misconduct. The BSB brings charges of professional misconduct; panels of the Bar Tribunals and Adjudication Service adjudicate them and decide sanction. An administrative sanction imposed by the BSB is not a disciplinary sanction and does not give a barrister a disciplinary record. Nonetheless, twenty open investigations indicates that this has become a live enforcement area rather than a theoretical risk.

What the courts said about chambers

The judicial dimension reinforces the collective point. The Divisional Court judgment in R (Ayinde) v London Borough of Haringey, heard together with Al-Haroun v Qatar National Bank, concerned judicial review grounds containing five authorities that did not exist. The barrister responsible was a pupil at the relevant time.

The court's observations went beyond the individual. It emphasised that there are serious implications for the administration of justice and public confidence if artificial intelligence is misused, and that measures must ensure every individual providing legal services in the jurisdiction understands and complies with their professional obligations and duties to the court when using AI. Practitioner commentary since has read that as imposing leadership responsibilities on chambers, extending to reviewing formal training on AI use and drafting policies and internal guidance.

That is the clearest signal available that the question is no longer only what an individual barrister did.

Formal rules may be coming

The current framework is guidance, not rules. That may not remain the position for long.

In its response to the Civil Justice Council's consultation on the use of AI in court documents, published on 21 April 2026, the Bar Council noted twice that the BSB aims to provide specific rules on AI use by barristers and is very likely to do so in the near future. The BSB's own 2026 to 2027 business plan separately commits the regulator to developing formal guidance on safe AI and technology adoption.

The CJC consultation itself closed on 14 April 2026 and its final recommendations are awaited. Among the proposals considered was whether legal representatives should be required to confirm that AI had not been used to generate or alter the substantive content of witness evidence. The Bar Council broadly supported requiring experts to identify AI use but raised concerns about embedding that requirement in the statement of truth, warning that a blanket declaration risks muddying the purpose of the statement of truth and may encourage tactical challenges over minor AI use.

For a set, the planning point is straightforward. Building a position now, while the framework is advisory, is considerably easier than retrofitting one to a deadline when rules arrive.

What chambers are actually doing

Sets that have moved have generally articulated principles rather than exhaustive rulebooks, which suits a membership of self-employed practitioners working across different practice areas with different risk profiles. A rule written for a crime practice may not fit a commercial one, and a document that tries to cover both tends to be ignored by everyone.

Reported examples give a sense of where the value has been found. Deka Chambers expanded its online library to include AI-assisted tools, with Stephen Glynn noting an improvement in the speed and quality of legal research offered to clients. At Cloisters, Robin Allen KC has argued that chambers should evaluate the benefits of AI against defined performance indicators, naming cost efficiency, accuracy, quality of work and productivity as the measures that matter. Both accounts date from 2025, and they remain among the few named, on-the-record examples from UK chambers rather than from vendors.

The pattern across the more thoughtful commentary is that successful adoption at the Bar has been narrow and specific rather than broad. Writing in Counsel in July 2026, Julie Ahadi described an emerging pattern of selective, highly targeted use by barristers, often small in scope but significant in impact, and argued that the goal should be meaningful rather than maximal adoption, with professional risk resting squarely on the individual barrister. The same piece noted that many chambers have responded by setting out clear principles rather than exhaustive rules.

What to measure

If a set is going to adopt anything, it should be able to say whether it worked. The natural starting points are the areas already under pressure: legal research, billing, document drafting and pricing. Measuring the effect on those is what distinguishes a decision from an enthusiasm.

One measure is non-negotiable regardless of what else a set tracks. The proportion of AI-suggested authorities that survive verification should be one hundred per cent on anything filed. Any figure below that is not an efficiency statistic, it is an incident report.

Building a chambers AI position

A workable position does not need to be long. It needs to answer questions that members will actually face, in terms that survive contact with a busy practice.

None of this displaces individual duties under the Core Duties and the Handbook. The purpose of a chambers position is to make compliance easier to achieve and easier to evidence, not to shift responsibility from the member to the set.

Questions worth asking a provider

Procurement is where the risk assessment the guidance contemplates actually happens, and the questions are not technical ones:

• Is material submitted to the system used to train models, and can that be switched off contractually rather than by a setting?
• What is the data retention period, and what happens to material on termination?
• Where is data stored and processed, and does that create any transfer issue?
• Who within the provider can access submitted material, and under what circumstances?
• Does the tool retrieve from an identified legal source, or generate from a general model? The distinction determines how likely fabricated citations are.
• What does the provider's own documentation say about accuracy, and does the contract disclaim it?

A provider unable to answer those in writing has told you something useful.

Supervision is the sharpest end of this

Of everything in the regulatory picture, supervision of junior practitioners is where chambers exposure is most concentrated.

A pupil asked to produce a note on a point of law, working to a deadline, with access to a general-purpose model, is in precisely the position that produced the facts of Ayinde. The failure mode is not laziness. It is that a fabricated authority produced by a competent model looks exactly like a real one: correct citation format, plausible court, plausible year, a proposition that sounds entirely orthodox. Nothing in the output signals that it is invented.

Experienced practitioners catch these because the case does not fit what they already know about the area. A pupil has no such background against which to notice the anomaly. That is why verification cannot be delegated to the person least able to perform it, and why a set's supervision arrangements, rather than its policy document, are what will determine whether it has a problem.

Measured adoption is a defensible position

It is worth saying plainly, because the surrounding commentary rarely does: the evidence does not show that the Bar is dangerously behind.

The regulator's own research found barristers risk-aware and pragmatic, using AI and task-specific tools for routine work while keeping human judgement central, and investing in new services at a higher rate than solicitors' firms. The guidance itself is enabling rather than restrictive. And the structural features that slow collective procurement at the Bar are real economic facts about the self-employed model, not symptoms of reluctance.

The realistic question for a set is not how much AI it can adopt, but which specific, bounded tasks it can improve while keeping the work defensible, and how it will know whether that worked. That is a slower answer than the market offers. On the current regulatory expectations, and on the evidence of what has gone wrong where it has gone wrong, it is the right one.

This article is information and analysis, not legal advice. Regulatory guidance changes, and the BSB has signalled that specific rules on AI use may follow. Confirm the current position with the BSB Handbook and the regulator's published guidance before relying on it.