Disciplined Integration: AI and the Legal Profession’s Existing Duties

Abstract

This article argues that the legal profession’s existing duties of competence, supervision, candour, confidentiality, and procedural fairness provide the necessary framework for integrating generative AI into legal practice, without requiring a wholly new regulatory paradigm. Narrowing its focus to transformer-based large language models, the article examines the ‘fake law’ phenomenon, where models generate plausible but non-existent authorities, and traces its implications for miscarriages of justice, particularly where adversarial safeguards are thinnest: self-represented litigants, resource-constrained parties, and off-record settlements. It then considers AI in adjudication, arguing that transparency, contestability, and human-authored reasons remain non-negotiable for judicial legitimacy. The central recommendation is disciplined augmentation, not automation: AI should operate within the profession’s existing load-bearing beams through auditable workflows, clear disclosure norms, and continuous education. Used with discipline, AI can strengthen practice; used without it, it risks eroding the trust and outcomes the profession exists to protect.