April 2026 · 5 min read
Saturday before last, Andrew Dietderich, co-head of the global restructuring group at Sullivan & Cromwell, sent an apology letter to Chief Bankruptcy Judge Martin Glenn of the Southern District of New York. An emergency motion his firm had filed in the Chapter 15 case of Prince Global Holdings was riddled with inaccuracies: fabricated case citations and references to legal sources that didn't exist. The errors, catalogued across three single-spaced pages of correction, were caught not by the firm but by opposing counsel.
The firm had policies on AI use. It had training. It had a manual citation-review process with multiple sets of eyes. In Dietderich's own explanation to the court, none of it caught the problem.
It is hard to read this filing without noting the obvious: Sullivan & Cromwell is the firm that advises OpenAI on the "safe and ethical deployment" of artificial intelligence. The firm literally helps write the playbook other institutions are supposed to follow. And it could not, on this specific motion, stop that same technology from introducing serious errors into a federal court filing.
Treat that as a data point, not a punchline. If the firm whose job is to think about AI safety for the largest AI company in the world can't reliably prevent hallucinations from reaching a judge, the baseline assumption about every other firm needs to shift accordingly.
Legal academic Damien Charlotin maintains a running catalogue of court cases in which AI-generated hallucinations have reached filings. That catalogue just crossed one thousand documented cases. The list spans solo practitioners, boutique litigation firms, and now the upper tier of the Am Law 100. The failure mode is not a function of resources, prestige, or training budget. It is a function of AI being used in legal work.
For insurance carriers, that's a meaningful distinction. A risk that correlates only with under-resourced firms is a risk that can be underwritten away with the right exclusions. A risk that shows up at every level of the profession, including at firms whose entire value proposition is rigor, is a risk carriers will treat as fundamentally difficult to diversify. That changes how they price it, and whether they price it at all on standard forms.
The shifts that follow are already underway.
The pace of AI exclusions attached at renewal will accelerate. The Verisk endorsements that came into circulation on January 1 (CG 40 47, CG 40 48, and CG 35 08) give carriers a clean way to carve AI out of general liability. That same logic is already migrating into professional liability forms, and the S&C filing gives underwriters a talking point they didn't have before.
Underwriting questions are getting sharper alongside it. For years, a professional liability submission has asked whether a firm has written AI policies. The S&C letter shows that the existence of a policy is not evidence of compliance with it. Expect carriers to move from "do you have a policy" to "describe your enforcement mechanism and the last three times it caught an error before filing." Firms that can't answer those questions credibly will see rate, retention, or both.
And affirmative AI coverage will stop being a niche product. The specialty markets offering AI-specific professional liability today (Testudo, Munich Re's aiSure, Armilla, Relm) exist because the standard market has decided AI risk isn't one it wants on its legacy forms. As more firms discover at renewal that their E&O has been quietly narrowed, demand for affirmative coverage moves from "interesting option" to "necessary purchase."
If Sullivan & Cromwell's safeguards didn't stop this, what makes yours different? That isn't rhetorical. It is the exact question a carrier will be asking at renewal, and the honest answer is what will determine whether a firm's AI exposure is covered, excluded, or priced in a way that forces a hard conversation with partners.
The firms that benefit from this moment will be the ones that treat the S&C filing as a preview of their own underwriting review rather than a story about someone else. That means pulling current policies and looking for AI-related endorsements: anything referencing artificial intelligence, machine learning, automated systems, or technology services. This means documenting how AI is actually used in legal work and where human review lives in the workflow. And it means having a conversation with a broker who understands that the answer to AI risk is not "more disclaimers in the engagement letter."
The letter to Judge Glenn was one firm's embarrassment. The next renewal cycle is every firm's problem.
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.