April 2026 · 4 min read
An AI hallucination is when a generative AI system produces output that sounds completely confident and is completely wrong. Not close enough. Not mostly right. The AI cites a case that doesn't exist, or it references a medical study that was never published. Sometimes it invents a drug interaction. Either way, the AI doesn't flag the error. It doesn't know there is one.
For lawyers and doctors, this has moved from theoretical risk to documented pattern, and the insurance industry is repricing accordingly.
Legal academic Damien Charlotin maintains the most comprehensive database of AI hallucinations in court filings. As of this month, the count stands at 1,334 documented cases globally, up from 719 in January. That's about five new documented cases per day. In March 2026, the Sixth Circuit sanctioned two Tennessee attorneys $15,000 each plus opposing counsel's fees for briefs containing more than two dozen fabricated citations.
And these are only the cases that were caught. Courts don't verify every citation in every filing.
In healthcare, hallucinations hide in the care pathway. A peer-reviewed 2024 study in the Journal of Medical Internet Research found GPT-4 hallucinating references at a 28.6% rate when asked to conduct medical literature reviews. A separate analysis published in Nature's Schizophrenia journal examined 115 AI-generated medical article references and found 47% were completely fabricated and another 46% were authentic but misrepresented, leaving only 7% accurate.
A fabricated case citation is sanctionable. A fabricated drug interaction or a confidently wrong clinical decision can injure a patient. And the liability question doesn't disappear because the mistake came from software.
Underwriters are already adjusting.
AI-specific exclusions are attaching to professional liability policies at renewal. Verisk's new 2026 endorsements (CG 40 47, CG 40 48, and CG 35 08) carved AI out of general liability, and similar logic is migrating into E&O and med-mal forms. W.R. Berkley introduced an "absolute" AI exclusion across professional and fiduciary lines.
Underwriting questions are also getting specific. "Do you have an AI policy?" is being replaced with "Describe your verification workflow and the last three times it caught an error." Firms and practices that can't answer credibly are seeing rate, retention, or both.
And affirmative AI coverage is moving from niche to necessary. Specialty carriers (Munich Re's aiSure, Armilla, Relm, Testudo) exist because the standard market has decided hallucination-driven losses are too concentrated to absorb on legacy forms.
Hope is not a coverage strategy. Neither is trust without verification. If your current policy doesn't have an affirmative AI endorsement (one that explicitly names AI-assisted work as covered work), then your position in a hallucination-driven claim is, at best, ambiguous. And ambiguity is where carriers win coverage disputes.
An AI endorsement names AI-assisted work as covered work product and tells you upfront what's covered and what isn't. Instead of discovering at claim time that your "ambiguous" policy was actually ambiguous against you.
Hallucinations will keep happening. The firms and practices that treat this as an underwriting problem to solve before renewal, not a PR problem to manage after claim, are the ones who'll be covered when it matters.
One last thing. If you can't explain exactly how an AI tool reached a specific recommendation (to a client, regulator, or jury) you can't catch its hallucinations before they cause a problem. That question, by the way, is question 6 on our coverage assessment. If you don't know the answer, your carrier is asking the same thing at renewal.
Sal Martorano is the founder of The AI Insurance Group, a licensed independent insurance agency in New Jersey, Pennsylvania and Florida. NJ Producer License No. 3004245927.