What is this course about?
Fabricated citations have produced sanctions, and the failure is mechanical rather than random. This course explains why language models generate plausible non-existent cases, builds a mandatory verification protocol, covers retrieval-grounded research against a real case database, and addresses supervision under Rules 5.1 and 5.3.
Who is this course for?
- Associates and staff attorneys
- Law librarians and research staff
- Supervising partners
- In-house counsel doing their own research
What do I need before starting?
- Legal research experience
- No technical background required
What will I be able to do afterwards?
- Explain mechanically why models fabricate citations
- Run a verification protocol that catches every fabrication
- Use retrieval-grounded research against a real case database
- Check whether a real case says what the summary claims
- Supervise associates and staff under Rules 5.1 and 5.3
What does each module cover?
Why do models invent cases?
35 minThe mechanism, explained clearly enough that lawyers stop treating it as a bug that will be fixed.
Objectives
- Explain generation versus retrieval
- Predict the conditions under which fabrication is most likely
- Stop expecting the problem to disappear with better models
Topics
Activity. Induce a fabricated citation deliberately and analyze what made it plausible.
What have courts actually done about it?
35 minThe sanctions record and the standing orders that have followed.
Objectives
- Summarize the sanctions record
- Identify what courts have required
- Track standing orders in your jurisdictions
Topics
Activity. Check for AI-related standing orders in every court you practise before.
What is the verification protocol?
45 minEvery citation, every time — the protocol and why partial verification fails.
Objectives
- Define a complete verification protocol
- Explain why sampling is insufficient
- Make the protocol efficient enough to survive deadlines
Topics
Activity. Verify every citation in an AI-assisted memo and time the process.
How does retrieval-grounded research differ?
45 minSearching a real case database rather than asking a model what it remembers.
Objectives
- Distinguish retrieval-grounded from generative research
- Configure research against a real database
- Recognize when the tool has left the database
Topics
Activity. Compare generative and retrieval-grounded research on the same question.
Does the real case say what the summary claims?
45 minThe second-order error — a real citation attached to a proposition the case does not support.
Objectives
- Verify the proposition, not just the citation's existence
- Detect mischaracterized holdings
- Check subsequent history
Topics
Activity. Check ten real citations for whether the case supports the stated proposition.
What must you disclose to a court?
35 minDisclosure obligations, certification, and what signing a filing now means.
Objectives
- Determine disclosure obligations by court
- Understand what certification covers
- Draft disclosure language where required
Topics
Activity. Draft AI disclosure language meeting the requirements of a real standing order.
How do you supervise under Rules 5.1 and 5.3?
40 minSupervisory duties when the work product came partly from a tool.
Objectives
- State supervisory duties under 5.1 and 5.3
- Design supervision that catches unverified work
- Establish accountability for verification
Topics
Activity. Design the supervision workflow for AI-assisted associate work.
Building the firm verification workflow
45 minThe workshop module: a workflow that makes unverified citations structurally impossible to file.
Objectives
- Build a workflow enforcing verification
- Make it non-bypassable under deadline
- Document verification for each filing
Topics
Activity. Build the workflow and attempt to file with an unverified citation.
What is the capstone project?
Firm verification protocol and workflow
Produce the firm's verification protocol covering citation existence and proposition accuracy, a supervision design under Rules 5.1 and 5.3, court disclosure language, and a workflow where filing with an unverified citation is structurally impossible.
Deliverable: A verification protocol and workflow adopted at firm or practice group level.
How are learners assessed?
- Every citation in a test memo verified for existence and proposition
- Workflow tested — filing with an unverified citation must fail
- Standing order check completed for every court the participant practises before
What ships with the course?
Facilitator guide
Session-by-session running order, discussion prompts, and the questions that reliably derail a room.
Learner workbook
Exercises, checklists, and the templates each module's activity produces.
Hands-on lab environment
A sandboxed ibl.ai deployment so exercises run against real agents, not screenshots.
Assessment bank
Scenario questions and rubric criteria mapped to each stated learning outcome.
Source bibliography
Every primary regulation and standard cited on this page, linked and dated.
Which AI agents does this course use?
The hands-on modules run against agents already deployable on the ibl.ai platform for legal.
Where does the course material come from?
Every module is grounded in primary sources — the regulation, standard, or research itself, not a summary of it. Each was resolved at authoring time.
- Model Rules of Professional Conduct
American Bar Association
Competence and supervision duties underpinning the protocol.
- CourtListener
Free Law Project
Primary case database used for verification exercises.
- Federal Rules of Civil Procedure Rule 26
Cornell Legal Information Institute
Procedural context for certification and disclosure obligations.
- Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks
Lewis et al., arXiv
Technical basis for the retrieval-grounded research in Module 4.
Delivery notes
Binding guidance for anyone preparing and delivering this course.
- Module 1's induced fabrication is the exercise that changes behavior. Attorneys who have generated a convincing fake citation themselves verify afterwards; those who have only read about it do not.
- Module 5 is the under-taught half. Verification programs check that cases exist and stop, while mischaracterized holdings are the more common and more dangerous error.
- Module 3 must time the protocol honestly. If verification takes longer than the research saved, say so and locate the actual efficiency gain elsewhere.
- Standing orders change constantly. Ship Module 2 as a process for checking rather than a list that will be stale within months.
- Do not name sanctioned attorneys gratuitously. The cases are public but the course should teach the mechanism, not create a spectacle.
Why run AI training on a platform you own?
You own the course, not a licence to it
Course content, learner data, and the platform run inside your perimeter — you own all the code and the data.
Model-agnostic delivery
Run the course's AI components on any LLM — Claude, GPT, Llama, Gemini, Command — and switch anytime.
No per-seat training licences
Usage-based or self-hosted, so cost tracks actual use rather than headcount.
Deploy anywhere
Cloud, private VPC, on-premise, or fully air-gapped — including for cohorts that cannot use public AI tools.
Frequently asked questions
What does the Verifying AI Legal Research: Never Cite a Hallucination course cover?
Fabricated citations have produced sanctions, and the failure is mechanical rather than random. This course explains why language models generate plausible non-existent cases, builds a mandatory verification protocol, covers retrieval-grounded research against a real case database, and addresses supervision under Rules 5.1 and 5.3. It runs 4.5 hours across 8 modules across 8 modules, at foundational level, and closes with a capstone: Firm verification protocol and workflow.
Who should take Verifying AI Legal Research: Never Cite a Hallucination?
It is written for Associates and staff attorneys, Law librarians and research staff, Supervising partners, In-house counsel doing their own research. Prerequisites: Legal research experience; No technical background required.
Can we run this course on our own infrastructure?
Yes. ibl.ai is model-agnostic and deploy-anywhere — cloud, private VPC, on-premise, or fully air-gapped — and you own all the code and the data. Cohort data, submissions, and any material learners upload stay inside your perimeter, which matters for legal teams that cannot send work to a public AI tool.
How do we get access to Verifying AI Legal Research: Never Cite a Hallucination?
Request access and we will set it up for your cohort — hosted by ibl.ai, or running against your own deployment. Tell us the group size and timing you need, and whether it should run inside your own perimeter.
How much does AI training for legal cost on ibl.ai?
There is no per-seat pricing — you pay for usage or self-host and pay only for the infrastructure, so a 5,000-person rollout does not cost 5,000 licences. 1.6M+ users across 400+ organizations run the platform this way, including NVIDIA, MIT, and Syracuse University.