This paper examines how AI is affecting professional practice in law, based on interviews with 44 US-based legal professionals. It looks at how attorneys are interpreting and responding to AI tools in their day-to-day work, set against a backdrop of both practical opportunities — such as faster document review at the discovery stage — and cautionary incidents, including cases where lawyers submitted court filings citing AI-generated fictitious case law. The research explores participants' direct experience of using AI in legal work, the factors helping or hindering adoption within firms, and the wider views lawyers hold about how AI might reshape the profession, its ethical obligations, and its institutions.

This essay challenges the assumption that AI will reduce demand for lawyers by drawing on the Jevons paradox, which states that efficiency gains tend to increase overall consumption rather than reduce it. It outlines four mechanisms by which AI-driven efficiency could instead expand the volume and scope of legal work, including its effect on unmet need among lower-income clients and its potential to create new categories of legal tasks. The paper also considers what this means for access to justice, legal education, and the possibility that regulatory responses to AI could recreate some of the compliance work the technology displaces.

This Stanford University paper examines the reliability of AI-powered legal research tools and, in particular, claims that specialist systems can avoid or eliminate the hallucinations associated with general-purpose AI models. The researchers focus on proprietary legal research tools from LexisNexis and Thomson Reuters, which use techniques such as retrieval-augmented generation (RAG) to draw on legal source material when responding to queries. Because these systems are closed, independently assessing their performance and the claims made for them can be difficult. The researchers systematically test these tools to assess how reliably they answer legal research questions. It develops a dataset and methodology for assessing their responses, including a framework for distinguishing hallucinations from accurate legal answers. The research also considers differences between systems in areas such as accuracy and responsiveness, and the wider implications for lawyers responsible for checking and supervising AI-generated legal research.

This paper looks at how well current rules and regulations - from international bodies down to professional conduct rules - actually cope with AI being used in legal work like research, drafting, contract review and predicting case outcomes. The author argues that present frameworks aren't built for the accountability, transparency, confidentiality, and access-to-justice questions AI raises, and sets out a model for how regulation could work better by combining formal law with professional standards and international coordination.

This paper examines how generative AI might affect legal conflict resolution. It draws on a survey in which members of the public and legal professionals across the EU and US wrote narratives about a future where AI is used throughout the legal process. The researchers analysed these narratives for recurring themes around risk and benefit, and for the legal tasks AI is expected to take on. They then compared responses from legal professionals against those from the public. They examined differences by regulatory context, contrasting the EU's AI Act framework with the US's more self-regulatory approach. The paper closes by setting out the resulting trade-offs facing legal sector decision-makers.

This paper examines the risks that generative AI tools may pose to client confidentiality when used by lawyers. It considers whether existing guidance identifies specific confidentiality risks arising from the way AI tools handle data, or treats the use of generative AI as a more general confidentiality concern. The paper proposes a framework for making that assessment. It separates confidentiality into three obligations: secrecy, security and loyalty to the client, and considers these alongside three types of risk associated with generative AI. It then examines ways of managing those risks and sets out a practical decision-making process to help lawyers determine whether a particular AI tool is suitable for their circumstances, requires additional safeguards, or should not be used.