The question he could not answer
A panel of New York's Appellate Division, Second Department, questioned a lawyer about the cases cited in his brief. Several did not exist. In his later submissions he said he had used traditional research alongside AI-assisted tools, could not recall which AI tool, and believed the invented citations came from the AI-assisted portion. The court described the brief as containing "fictitious purported Court of Appeals quotations wholly contrary to actual law" and imposed $10,500 in sanctions: $8,000 on the lawyer and $2,500 on his firm.
The policy existed
His firm required anyone using AI to personally review and confirm what it produced, and he acknowledged breaching that requirement. It is the policy most organizations have written. It is sound as a rule, and it runs entirely on one individual's diligence, under deadline, every single time.
Two failures with one root
The review did not happen, and then the record of what AI had contributed was missing too. Both depended on the same person's memory, so both failed at once. Reminding staff to be careful, or adding an expectation that they log their AI use, loads more weight onto that same single point.
Designing the dependency out
Provenance should come from systems, not recollection. When AI work happens in tools that retain their own history, the prompts and responses can be produced on request by anyone with access. Verification should be a state in the workflow rather than a personal virtue: a piece of work is not marked ready until a second reviewer has seen the sources behind its load-bearing claims.
For a leadership team, the diagnostic is a single question about any recent deliverable. If its author were unreachable tomorrow, could anyone else establish what AI contributed to it? Where the answer is no, the gap is organizational, and it is far cheaper to close before someone outside the building asks.
The assessment maps where your AI practices depend on individuals: See where your AI practices rest on one person →
Keep reading: Ask when they last found something · Understanding AI Hallucinations
This is the thinking behind a Trusted AI Culture.
Sources: New York Daily Record, "Attorney and law firm sanctioned for AI mistakes in court filing," June 26, 2026. Minnesota Lawyer, "Attorney, law firm sanctioned for AI mistakes in court filing," July 2, 2026.
Written with AI in the loop: my idea, AI drafted and sharpened, my judgment on the way out. Every word is mine to stand behind. — Darren