AI Safety & Society · The 2026 incident files
You can describe the hallucinated-case-law phenomenon — why it keeps happening, who is being sanctioned, and what verification it forces on professional users.
The cleanest deployment failure of the LLM era is also the funniest, until you read it twice. Lawyers keep filing court briefs containing citations to cases that do not exist. The model invented them. The lawyer did not check.
The pattern is no longer anecdotal because courts now write it down in opinions. Mata v. Avianca named the failure in 2023: a lawyer treated ChatGPT like a legal search engine, filed fabricated authorities, and asked the same model to vouch for them. Prososki v. Regan shows the next civic consequence: a brief stricken, an appeal dismissed, and counsel referred for discipline. Couvrette v. Wisnovsky shows the workflow consequence: local counsel cannot be a signature-only rubber stamp while another lawyer files fake cases and fabricated quotations.
This chapter walks the arc from Mata to later court records. The civic question is what kind of failure this is. It is not only a model bug. It is a deployment-discipline failure that no amount of model improvement can fix unless the professional workflow includes verification.
Type: multi-choice
Prompt: > What does the hallucinated-case-law pattern primarily expose?
Choices: - (a) A deployment-discipline failure: professionals submitting AI output without verification (correct) - (b) A bug that newer models have fully fixed - (c) A flaw unique to one chatbot product - (d) A problem that only affects junior lawyers
Chapter contains 3 lessons.