Since 2023, there have been widely reported cases of lawyers who filed briefs containing case citations produced by a generative AI tool that turned out not to exist. Courts responded with sanctions, orders and, in some jurisdictions, standing rules about AI use. The underlying problem is not new technology so much as old professional responsibility: a lawyer must verify what goes out under their name.
This post explains why AI fabricates citations and sets out a workflow that makes verification routine.
Why AI Invents Cases
Large language models predict plausible text. They do not look up cases in a database unless they are specifically connected to one. When asked for authority, a general-purpose chatbot may produce a citation with a realistic party name, reporter, volume and page, and a convincing summary, none of which is real. The format is easy to imitate. The substance is not guaranteed.
Even tools built for legal research that retrieve real documents can misstate holdings, cite the wrong proposition or blur jurisdictions. Fabrication is the most dramatic failure, but subtle error is more common.
What the ABA Has Said
ABA Formal Opinion 512, issued in July 2024, discusses generative AI in light of the Model Rules. As the opinion describes it, lawyers using these tools should understand their capabilities and limitations under Rule 1.1, and should independently review output. The degree of review depends on the task and the tool. Rules 5.1 and 5.3 on supervision also apply to how junior lawyers and staff use AI. Your state bar and individual courts may add their own guidance or requirements, so check them.
A Verification Workflow
1. Treat AI output as a lead, never as authority
Use AI to brainstorm search terms, outline arguments or summarize documents you already have. Do not cite a case you have not read.
2. Pull every citation from a trusted source
For each case, statute or rule cited:
- Find it in a reliable legal research database, not through the AI tool.
- Confirm the case exists, the citation is correct and the court is right.
- Read the relevant portion of the opinion.
- Confirm it says what the brief claims it says, including quotations.
- Check subsequent history with a citator to see whether it has been overruled or limited.
3. Verify quotations and pin cites
AI may paraphrase and present the result as a quote. Compare quoted language against the source, word for word.
4. Record the check
Keep a short log or cite-check table in the matter file noting who verified each citation and when. This helps with supervision and with responding to any later question.
5. Require a second reader on court filings
For significant filings, have someone other than the drafter spot-check citations.
Policy Elements
- State plainly that no AI-generated citation may be used without independent verification.
- Identify which tools are approved for research and which are not.
- Require disclosure internally when AI assisted a draft, so reviewers know to look harder.
- Check local rules and standing orders for AI disclosure or certification requirements in each court where you file.
- Train staff with real examples of how plausible fabrications look.
Warning Signs in Output
- Citations that are unusually perfect for the argument you wanted
- Cases you cannot find in any database
- Mismatched courts, years or reporters
- Quotes that do not appear in the opinion
- Confident answers about obscure or very recent law
Protect Confidential Data Too
Verification solves accuracy, not confidentiality. Do not paste client information into unapproved tools. Use enterprise tools with appropriate contractual protections for client work, and keep your AI policy current.
Culture
Make it safe for associates and staff to say "I could not verify this." The pressure of a deadline is when shortcuts tempt people most.
Our Role
Counsel Cyber helps firms select and configure AI tools, control what data can flow into them and write policies that include verification requirements. If you are drafting an AI policy, we are glad to help.