Where Should Courts Draw the Line on AI-Assisted Filings?

Summary

This opinion article argues that using AI in legal work is generally acceptable, but signing or submitting AI-generated material without reading and verifying it is not. It cites a database containing at least 2,000 legal cases involving AI hallucinations, with the United States accounting for 1,473 cases, followed by Canada, Australia and the United Kingdom. Examples include fabricated cases and quotations that led to a $110,000 penalty, a Nebraska suspension after 57 of 63 citations were defective, sanctions against all four lawyers in a case involving hallucinated citations, a UK lawyer being struck off, and an Arizona sentence being vacated after an AI-generated video of a deceased victim was used at sentencing. The article says co-signing can count as AI use, while courts have generally treated pro se litigants more leniently than lawyers. It also describes AI-related errors or concerns involving judges, court staff, the Justice Department and HHS materials. Across jurisdictions, emerging rules tend to require signers to certify that authorities are genuine and accurately cited, or that AI-assisted work received independent human verification. New York permits AI without disclosure if the filing is reviewed; Florida replaced disclosure orders with a statewide certification rule; California’s 2027 law requires disclosure, personal verification of citations and protection of confidential data; and an Iowa rule scheduled for 2027 would bar judges from delegating decision-making to AI. The article identifies drafting, editing, summarizing and research as generally acceptable when checked by a human, while unread citations, invented facts or testimony, and concealed prompts cross the line. Its conclusion is that accountability should attach to the person who files the material, and that eliminating AI from legal work altogether would be an overreaction.