Toolkit / Best Practices for AI in Legal Education
Best Practices for AI in Legal Education
Fall 2026 principles for responsible AI use in legal education: course policies, learning goals, student use, faculty use, verification, confidentiality, and equity.
This guide gives the durable principles behind the Project’s teaching guidance. For copyable syllabus and assignment language, start with the AI Syllabus Guide. For Penn-specific tool access and data-policy details, use the AI Resources Portal.
The recommendations here are not Law School policy. Faculty set the rules for their own courses, and students should follow the syllabus, assignment instructions, exam rules, clinic rules, journal rules, and supervisor instructions that apply to their work.
Start with the course rule
A course AI policy should answer the questions students actually face:
- May I use AI to study or prepare for class?
- May I upload course materials, slides, notes, recordings, or exams?
- May I use AI for brainstorming, research leads, outlining, drafting, revising, citation checks, or editing?
- Must I disclose AI use?
- Which assignment rule controls if the syllabus and assignment instructions differ?
One sentence rarely does the job. Use a course-level default, then add assignment-specific instructions when the task changes. The key is to regulate the work, not the technology in the abstract.
Student use by task
AI use depends on what the student is trying to do.
| Task | Good use | Caution |
|---|---|---|
| Class preparation | Ask for practice questions, compare an outline to class notes, or get a plain-language explanation after doing the reading. | Do not replace reading, briefing, or synthesis with AI summaries. |
| Brainstorming | Generate possible paper topics, counterarguments, or research paths. | Treat output as a starting list, not as authority or analysis. |
| Research | Ask for leads, search terms, or categories of sources. | Verify every case, statute, quotation, and factual claim against the source itself. |
| Drafting | Use only if the course or assignment permits it. | Undisclosed AI-generated text may defeat the assignment and raise academic-integrity concerns. |
| Revision | Ask for clarity, organization, tone, or counterargument feedback on a student-written draft. | The student remains responsible for substance and must disclose use when required. |
| Study practice | Generate hypos, flashcards, multiple-choice questions, and oral-argument practice prompts. | Check answers against assigned materials and the professor’s coverage. |
Faculty use by course type
Faculty choices should track the course design.
Traditional lecture courses. Separate AI study use from exam use. Students may benefit from AI-generated practice questions and explanations, but the exam rule should be explicit and usually belongs in the exam instructions as well as the syllabus.
Seminars with research papers. Use staged process requirements: topic memo, source list, thesis paragraph, outline, partial draft, conference, final paper. Decide which stages permit AI and require disclosure when AI shapes the work.
Writing, clinics, and skills courses. Tie the AI rule to the skill being assessed. If the task tests first-pass legal analysis, restrict drafting. If the task tests revision, client communication, or professional judgment, AI can become something students critique and supervise.
Mixed-assessment courses. Use layered rules. A course may prohibit AI on quizzes, permit study use, require disclosure on practice problems, and integrate AI into one assignment.
Design choices that reduce misuse
Good assignment design does more than announce a rule.
- Use process evidence. Ask for outlines, research trails, source lists, revision memos, or short oral explanations.
- Make AI an object of critique. Have students evaluate an AI answer, find hallucinated authority, improve a weak prompt, or compare AI output with a model answer.
- Use course-specific facts. Tie assignments to class discussion, assigned materials, local hypotheticals, or current course themes.
- Require source verification. AI output is a lead, not authority.
- Use low-stakes practice. Let students practice supervising AI before a high-stakes assessment.
Do not build academic-integrity work around AI detection. Performance varies by system and context, and a detector score is not standalone proof of misconduct. Detection also does not teach students how to supervise AI responsibly.
Accuracy, attribution, and source checking
AI systems produce fluent text whether the output is right or wrong. That is the central risk for legal work.
Students and lawyers should assume any legal authority, quotation, statistic, factual claim, or citation surfaced by AI is unverified until checked against an authoritative source. Legal-specific systems reduce some risk by retrieving from legal databases, but they do not eliminate the need for human review.
Disclosure and citation serve different functions. Disclosure tells the professor how AI affected the process. Citation tells the reader where an idea, quotation, source, or authority comes from. In writing-intensive settings, students may need both.
Privacy and confidentiality
AI policy is also data policy. Do not upload student work, grading materials, class recordings, exam materials, client information, confidential facts, nonpublic committee work, or other sensitive material unless the tool is approved for that material and the course or supervisor authorizes the use.
The current Penn-specific tool and data-handling rules live in the AI Resources Portal. When in doubt, use that portal rather than relying on a copied account or privacy claim in a syllabus or guide.
Access and equity
AI requirements can create access problems. Some tools cost money, some tools are available only to particular user groups, and some students will have more experience than others.
If AI use is required, faculty should specify the tool or tool category, make sure students can access it, provide an alternative when needed, and explain what role AI work will play in grading. If AI use is optional, students should understand that they will not be penalized for choosing not to use it.
Glossary
Generative AI — software that generates text, images, code, audio, or other content in response to prompts.
Large language model (LLM) — a model trained on large text collections to predict and generate language. ChatGPT, Claude, and Gemini are examples of interfaces built around LLMs.
Hallucination — a confident but false output, including invented cases, citations, quotations, facts, or sources.
Prompting — the practice of giving instructions, context, examples, and constraints to shape AI output.
Retrieval-augmented generation (RAG) — a system design that asks the model to answer using retrieved source materials. RAG can improve grounding, but the user still must verify the answer.
AI-use statement — a short disclosure explaining which tool was used, for what purpose, at what stage, and how the user checked the output.
Related Toolkit guides
- AI Syllabus Guide — copyable policy and assignment language.
- AI Use Policy Templates — attribution and disclosure examples for writing-intensive seminars.
- Teaching Examples and AI Demos — reusable teaching prompts and activities.
- Prompt Guide & Scenarios — prompting patterns for legal education.
- AI Resources Portal — current Penn-specific tool access and setup guidance.
Acknowledgment
With thanks to the AI Law Lab alumni who helped create the original version:
- Meghana Bhimarao ‘25 — AI Law Lab & CTIC Fellow
- Lakshmi Prakash ‘25 — AI Law Lab & CTIC Fellow
Status
Maintained for the Penn Carey Law community.