The draft read perfectly. The cases never existed.
You read the AI's draft and your gut said yes. The note was clean, the brief cited real-looking cases, the analysis hit every beat you would have hit. So you signed it. That is exactly the moment this book is about.
The most dangerous AI error in your field is not the obvious blunder a junior would catch. It is the confident, polished, mostly-correct answer that sails past a busy expert precisely when scrutiny should rise. A real attorney once filed a federal brief built on cases that never existed, generated and quoted so convincingly that a seasoned professional signed his name to them. In medicine, law, and science, fluency now masquerades as competence, and the same trap is set in every differential diagnosis, legal research memo, and reconciled schedule you touch.
The Second Opinion on AI reframes your core professional skill from operating AI to verifying it. Treat every output like a sharp junior colleague's draft: a single read that ships nothing until it clears an independent check.
A working method for verifying AI in high-stakes work:
- Spot the confident-wrong answer, the fluent, mostly-correct output that slips past a busy expert because it reads like your own best work.
- Tell in seconds which outputs are safe to sign and which must clear an independent check before they touch a patient, a filing, or a published result.
- Catch fabricated citations and phantom studies before a real-looking case name or reference goes out under your signature.
- Trust calibration for AI outputs: replace gut feel with a repeatable read on when to rely on the machine, when to doubt it, and how to prove your call.
- The second opinion as a working method: nothing ships on a single read, whether you are a physician charting, an attorney drafting, or an auditor fact checking a schedule.
Read it and you will know, before you sign, whether the draft in front of you deserves your name. Capability is now assumed; every professional in your field can call the same models. The scarce, defensible expertise, the one no model upgrade erases, is knowing when to trust AI, when not to, and how to prove it. Your license, your bar card, and your good name depend on that judgment.
For physicians, attorneys, auditors, scientists, and the risk and compliance professionals who govern them. Part of the Lead, Govern & Apply AI series, in The Verifier's Library.
