Many will agree with AI skeptics that lawyers should not use AI in criminal cases. My take is different.
Those lawyers are right about one thing: criminal law deserves the strictest checking discipline in the profession. That is not a prescription for AI abstinence.
The real problem is lawyers filing something without reading it. That sin predates AI by a few centuries.
Suppose a lawyer asks a new law clerk to review his draft memo, then submits it to the court without reading it. How many would conclude from this experience that law clerks are useless?
Hallucination is a genuine risk in legal research. Avoiding them takes work, but it’s not a catastrophic risk when a lawyer knows how to use AI. Nothing the AI model says goes into the brief unless the lawyer includes it.
The reality is that prosecutors, defense counsel, and even judges who know how to use these tools have a large advantage over those who don’t.
Skeptical? Try this. Take a brief you’re proud of, one in your strongest area. Feed it to a frontier model with a one-word prompt: “Critique.”
Every time I run this test, even on subjects I’ve worked with for thirty years, the AI app finds something I missed.
Every time.