Counsel in the Age of AI

This summary, prepared by TrialView, reflects the themes, debates and questions that emerged from a closed roundtable discussion held under the Chatham House Rule. No statements are attributed to any individual participant or organisation. The discussion drew on perspectives from across the legal profession, including the Bar, the arbitration sector, in-house practice, private practice, and the legal technology industry.

Overview

The roundtable, hosted by 4-5 Gray’s Inn and TrialView, brought together a cross-section of the legal profession to examine how artificial intelligence is reshaping legal practice, and crucially, where it is not yet doing so fast enough. Across all areas of discussion, a set of interrelated tensions emerged; between transparency and accountability; between institutional adoption and individual autonomy; between the pace of technological change and the pace of regulatory response; and between the promise of AI and the profession's capacity to assess that promise critically.

Not all agenda items were addressed in full. The discussion was deliberately discursive, and several themes recurred across different parts of the session, reflecting the degree to which the profession's AI challenges are interconnected rather than siloed.

Transparency: A Central but Contested Imperative

Transparency from legal technology providers

A strong consensus emerged that legal technology providers are not yet being held to adequate standards of transparency. Participants noted that benchmarking claims made by AI vendors are frequently difficult to interrogate, that accuracy data is rarely disclosed in a form that practitioners can meaningfully evaluate, and that the gap between marketing and demonstrable performance remains wide.

The view expressed was that providers operating in the legal market should be required, whether through regulation, professional standards, or procurement expectations, to disclose how their tools have been tested, against what datasets, and with what results. Without this, practitioners are being asked to accept risk without the information needed to assess it.

Transparency obligations across the profession

Participants agreed that transparency about AI use should extend across all segments of the profession, but that what this means in practice will vary considerably depending on role and context. The risk profiles of a barrister appearing before a court, an arbitrator making procedural decisions, an in-house counsel advising on transactions, and a trainee drafting research notes are not the same, and a single disclosure standard is unlikely to be fit for purpose across all of them.

The question of how granular disclosure obligations should be, and to whom,  remains unresolved. Discussions ranged from whether parties to litigation should be required to disclose the use of AI tools in the conduct of proceedings, to whether judicial members and arbitrators should operate under different, more stringent requirements. No consensus was reached, but the direction of travel was clear: the profession cannot continue to operate in the absence of defined standards.

Transparency, the EU AI Act, and regulatory coherence

The EU AI Act introduces transparency and accountability obligations that apply across industries, including the legal sector. Participants noted the tension between a horizontal regulatory framework designed for broad application and the need for profession-specific disclosure norms that reflect the particular context of legal proceedings. How these frameworks interact, and whether domestic professional regulation will be sufficient to fill the gaps the EU AI Act leaves, was identified as an open question requiring active engagement from the professional bodies, and beyond.

Balancing transparency with personal accountability

A recurring theme was the risk that disclosure frameworks, if poorly designed, could shift accountability in ways that are neither fair nor functional. Transparency should not become a mechanism for deflecting professional responsibility onto tools or third-party providers. Participants emphasised that the duty of competence, the duty of candour, and the obligations owed to clients and to the court remain personal obligations, and that any framework for AI transparency must reinforce rather than dilute that principle.

The Education Gap

The education deficit was identified as the foundational problem underlying almost every other issue discussed. It is not simply a question of training practitioners to use AI tools correctly; it is a question of equipping the profession to understand what AI is, how the tools they are using or encountering actually work, when use is appropriate, and what the limits and failure modes of those tools are.

Specific concerns raised included:

• Practitioners, including experienced counsel, frequently do not understand how large language models generate outputs, and therefore cannot reliably identify hallucinations or assess the reliability of AI-assisted work product.
• There is limited understanding across the profession of what AI vendors mean by accuracy, reliability, or validation, making it difficult to assess competing claims or push back on inadequate disclosure.
• The question of when and what to disclose about AI use is genuinely unclear to many practitioners, with guidance lagging well behind practice.
• The absence of training is driving shadow AI use, i.e., practitioners reaching for accessible consumer tools without governance, oversight, or any framework for assessing risk.

Who owns the education piece was not resolved. Views were expressed that the responsibility falls variously on regulators, professional bodies, chambers, firms, and individual practitioners. The risk of fragmented or duplicative guidance frameworks, with the Bar Council, the Law Society, and the Judiciary each developing separate approaches, was noted, as was the risk of no coordinated approach at all.

The Junior Lawyer Problem

One of the more striking themes to emerge was concern about the impact of AI on the development of junior lawyers, trainees, and students. There was recognition, expressed with varying degrees of alarm, that a significant cohort of early-career practitioners are already outsourcing core cognitive tasks to AI: research, drafting, analysis, and in some cases reasoning itself.

The question posed, and not fully answered, was whether this represents the most significant long-term risk of AI adoption in the profession. A generation of lawyers who have not developed the foundational skills of legal reasoning, research, and drafting independently may not be in a position to supervise, validate, or critically assess AI outputs, thus creating a compounding problem as these practitioners progress and take on more senior roles.

This is not, participants noted, a reason to prohibit AI use by junior lawyers. But it does suggest an urgent need for supervisory frameworks, training curricula, and professional development standards that ensure AI is used as a tool to augment learning rather than to replace it.

Arbitration, Litigation and the Question of Context

A distinction was drawn between the use of AI in arbitration and in litigation, with participants noting that arbitrators may operate with greater flexibility in how they manage proceedings and what forms of assistance are permissible. The analogy of a tribunal secretary was raised in the context of certain AI-assisted functions (drafting procedural orders, summarising submissions, organising hearing materials) where AI assistance might be considered broadly analogous to existing practice.

This contrasts with the position in court proceedings, where the rules of evidence, disclosure obligations, and judicial accountability create a different and more constrained environment. The point was made that the profession may need context-specific norms rather than a single cross-cutting framework, and that arbitration's relative flexibility could serve as a useful testing ground for developing good practice.

Witness Statements and AI Drafting

The use of AI in the preparation of witness statements was discussed, with participants acknowledging the genuine risks, including the potential for AI-generated content to introduce inaccuracies, flatten witness voice, or obscure the distinction between what a witness recalls and what has been drafted on their behalf.

However, it was also noted that the profession has long operated with conventions around solicitor-drafted witness statements, and that the boundary between acceptable drafting assistance and impermissible authorship is not new. The question is whether AI changes the nature or degree of that risk in ways that require a distinct regulatory response, or whether existing professional conduct obligations, when properly applied and enforced,  are sufficient. No firm conclusion was reached, but the issue was identified as warranting specific guidance.

AI in Disclosure: Emerging Conduct Risks

The use of AI in document review and e-disclosure was identified as one of the areas of most immediate and acute professional conduct risk. A concern was raised that disclosure of AI prompts is already being used as a tactical lever in proceedings: parties being pressured to settle on unfavourable terms rather than face the requirement to disclose how AI tools were used in the conduct of the case.

This dynamic, if it becomes established practice, raises serious questions about the integrity of proceedings and the ability of parties to use AI tools without creating collateral vulnerabilities. Participants noted that professional guidance on AI use in disclosure is substantially behind practice, and that the question of where the duty of competence sits remains unresolved.

In-House Counsel: The Unrealised Promise of Efficiency

There was a candid acknowledgement that in-house counsel continues to await the cost savings that AI adoption was predicted to deliver. The gap between the efficiency gains promised by vendors and the experience of in-house teams (many of whom remain in early or exploratory stages of adoption) was noted.

Contributing factors discussed included the complexity of integrating AI tools into existing workflows, concerns about data security and privilege, the absence of clear governance frameworks, and the reality that the most resource-intensive aspects of in-house legal work do not always lend themselves readily to AI-assisted processing. The expectation management question was left open.

The Sustainability Question

A question was raised that is rarely part of mainstream legal technology discourse: what happens if AI becomes economically or environmentally unsustainable? The energy demands of large language models, the cost structures of frontier AI providers, and the concentration of infrastructure in a small number of technology companies were identified as systemic dependencies that the profession has not seriously interrogated.

The point was not to suggest that AI adoption is likely to collapse, but to note that a profession that has restructured workflows, reoriented training, and reorganised its commercial models around AI dependency has made itself vulnerable in ways it may not have considered. This was offered as a prompt for reflection rather than a prediction, but it was received as a legitimate and underexamined risk.

Themes and Areas for Further Work

The following themes were identified as warranting further attention, whether through follow-up working groups, regulatory engagement, or sector-specific guidance development:

• Standards for AI vendor transparency and benchmarking disclosure in legal technology procurement
• Profession-specific AI disclosure frameworks that reflect different risk profiles across the Bar, the judiciary, solicitors, and in-house counsel
• The interaction between the EU AI Act and domestic professional regulation, and where the gaps lie
• Training and education frameworks for junior lawyers, with supervisory obligations that ensure AI augments rather than replaces skills development
• Professional conduct guidance on AI use in disclosure and witness statement preparation
• Context-specific norms for AI use in arbitration, developed in dialogue with arbitral institutions
• The long-term systemic dependencies created by AI adoption, including energy, cost, and infrastructure concentration