Forging ahead or proceeding with caution: Developing policy for generative artificial intelligence in legal education

Synthesis: Gutowski and Hurley survey how ABA-approved US law schools govern generative AI and find a sector making policy under time pressure without an evidence base to make it with. Their framework scores institutional policies on five dimensions — prohibitiveness, permissiveness, educational integration, transparency and accountability, and depth — each on a 0–5 rubric, and the canvass shows most schools taking generally prohibitive positions while reserving discretion to individual instructors and committing to revisit the rules as the technology moves. The article is a law review policy analysis rather than an empirical study, but it reports the sector's own data honestly, including that the ABA's 2024 survey drew responses from only about 15% of accredited schools. Its substantive contribution to the knowledge base is the argument that legal education must govern AI as preparation for a regulated profession: the professional conduct rules that bind practising lawyers already apply to the tools, which makes clarity, training and flexible AI Governance matters of assessment validity rather than compliance alone.

Key Findings

1. Five dimensions, scored on rubrics. Policies are assessed for prohibitiveness (0 = no restrictions mentioned, 5 = general prohibition on all use), permissiveness (0 = no allowance, 5 = full encouragement with comprehensive support), educational integration (how far AI is built into the curriculum), transparency and accountability (clarity of reporting, citation and enforcement), and depth (comprehensiveness of the policy text). The permissiveness and prohibitiveness ratings correlated inversely, which the authors treat as the expected result of two sides of one stance.
2. Most law schools were prohibitive at the time of writing. The dominant pattern is a generally restrictive policy, with institutions explaining the stance as a pause: rapid Large Language Models (LLMs) growth left administrations and faculties needing time to understand the technology. Crucially, nearly all prohibitive policies also allow individual instructors to permit use in their own courses, and most include language committing to review and revision as capabilities change.
3. There is no accepted approach, and coverage is uneven. Policies run from comprehensive governance frameworks to no stated policy at all, shaped by each school's mission and governance culture. The authors present this diversity as evidence of the problem's complexity rather than as a ranking to be resolved.
4. The sector's evidence base is thin. The ABA's 2024 AI and legal education survey drew responses from only 29 schools, roughly 15% of ABA-approved law schools, so its patterns cannot be treated as representative. A LexisNexis survey of 800 law students found only 9% reporting current use of generative AI for their studies, with 25% intending to adopt it. Low reported use, the authors suggest, is partly a response to unclear or absent policy and the resulting fear of violating rules nobody has explained.
5. Hallucinated authority is the discipline's signature failure. They cite Dahl et al.'s finding that when asked a direct, verifiable question about a randomly selected federal court case, LLMs hallucinate between 58% (ChatGPT-4) and 88% (Llama 2) of the time, alongside reports of practitioners sanctioned for filing fabricated case law. Their treatment is that such incidents argue for training and supervision, not for abandoning the tools.
6. Generative AI passes the licensure gates. The models have passed both the Bar Exam and the Multistate Professional Responsibility Exam, which the authors read as exposing the limits of a minimum-competency barrier rather than as evidence of genuine legal understanding.

How the study was conducted

The article is a doctrinal and policy analysis rather than an experiment. The authors canvass institutional generative AI policies across ABA-approved law schools, score them against the five rubrics, and then work through the underlying considerations: academic integrity and professional ethics, permissive versus prohibitive stances, and faculty impact, disclosure, citation and alternative requirements. They present model policies at each rubric level as illustrations of what a given score looks like in practice, drawing on published policies from named institutions including Harvard, the University of Chicago, Temple, Loyola Chicago, USC, NYU and the University of Edinburgh, and they discuss the practical realities of policy creation in law schools, where faculty hold unusual individual authority over curriculum and standards.

Why legal education is a hard case

Three features distinguish it from generic higher education governance.

• Faculty governance. Shared governance gives law faculty a direct and often individual say over course standards, so policy has to be built with faculty rather than announced to them. The authors treat this as an added complication absent from most institutional AI policy.
• Clinical education resists the tools on pedagogical grounds. Reporting the position taken by Karr and Schultz, they argue that generative AI is not compatible with what clinical legal education is trying to achieve and should therefore not be used with students there: tools designed to mimic human responses do not develop original thought, understanding, or the client-interaction and ethical decision-making skills clinics exist to build.
• Professional duties arrive with the graduate. The ABA's duty of technological competence in the Model Rules of Professional Conduct, and its formal guidance on generative AI covering confidentiality, communication with clients, meritorious claims, candour toward the tribunal, supervision of others using the technology, and reasonable fees, mean the school's rules sit downstream of rules the student will be bound by. Law schools, on the authors' account, must ensure technological competence for their students just as continuing legal education must now keep practitioners current.

Considerations the authors identify

• Independent thought and fraud. Students must engage in original thinking; the risk they name is passive acceptance of model output without evaluating sources or building legal reasoning. Grading is meant to assess individual comprehension, not the quality of a model's recall, so over-reliance compromises the measurement that licensure and the profession assume.
• Disclosure, citation and alternative requirements. The article reports disagreement about whether any use must be disclosed, examples of disclosure forms students file with instructors, and no consensus on how scholarly or class writing should cite AI, noting that the Bluebook currently has no guidance. They also make a prediction that cuts against the rules being drafted now: at some point AI integration will make disclosure as pointless as noting that a student used a search engine.
• Assessment design as the alternative to prohibition. When AI can do the task, the authors' preferred response is a task it cannot do easily: redesigned assignments requiring nuanced legal analysis or multi-step Problem Solving that resists a single generated answer.
• Instructor discretion. Policies should let instructors decide per course, because the professor is best placed to judge whether AI use undermines the competency a given assessment is meant to measure. Their example is permitting AI for legal research while prohibiting it in a writing assignment whose object is advanced analysis.

Implications for policy and practice

The recommendations are procedural and deliberately modest: guidelines that are clear and comprehensive regardless of institutional stance; full involvement of students, faculty and staff in drafting, both to prevent backlash and to produce well-rounded rules; policymakers well informed about impact, enforceability and adaptability before they act; proactive training for students and faculty; governance designed to be flexible because the technology changes continuously; and periodic review, on the authors' view that policy generation is not a one-time event. Their closing position is institutional self-regulation: schools should not wait for the ABA to dictate policy but should share what works and govern collaboratively. They frame the whole exercise as a call to action and a commitment to integrity, while conceding that reasonable minds can differ on the substance.

Limitations

The article is a policy analysis, not an evaluation of learning outcomes: it does not measure whether permissive or prohibitive stances affect student learning, bar passage or practice readiness, and its policy canvass is a snapshot that the authors themselves expect to date quickly. The survey figures it relies on are limited by low response rates and self-report, and the model policies it holds up as exemplars are selected illustrations rather than a random sample, so the framework is best read as a governance instrument for institutions rather than as the sector's measured state.

Connected Concepts

• Academic Integrity — the conduct framework the policies formalise
• Assessment Validity — whether assisted work still measures legal competence
• Authentic Assessment — redesigning tasks that AI cannot complete for the student
• AI Use and Disclosure Statements — the disclosure and citation norms still without consensus
• AI Governance — faculty governance as the constraint on law school policy
• Educational AI Policy — institutional AI policy compared across a sector
• Critical Thinking — original legal reasoning as the learning objective at risk
• Socratic Method — preparation-dependent dialogue under Socratic teaching
• Experiential Learning — clinics, writing courses and practice-based training
• Workplace Learning — the licensed professions family this discipline belongs to
• Career Development and Readiness — practice readiness as the stated purpose of policy
• Hallucination Risk — fabricated citations and their consequences in practice
• Legal Education — the discipline page this article anchors
• Legal Issues and Risks — the liability questions institutions face when governance is unclear

Connected Articles

• Governing generative AI in higher education: a global Delphi study on policy and practice — Global Delphi study on generative AI policy and practice in higher education
• Policy Fragmentation or Institutional Alignment? Institutional Governance of AI in Universities and Business Schools — Whether institutional AI governance aligns or fragments across units
• Artificial Intelligence in UK Higher Educational Policy and Institutional Decision Making — UK higher education policy and institutional decision making on AI
• The Integrity of Psychology Assessments in the AI Age: A Critical Examination — Assessment redesign and the pass boundary in a professional programme

Citation

Gutowski, N. N., & Hurley, J. W. (2025). Forging ahead or proceeding with caution: Developing policy for generative artificial intelligence in legal education. University of Louisville Law Review, 63(3), 581.