Policyholder Pulse
“I continue to believe—perhaps more so with each interaction—that LLMs have something to contribute to the ordinary-meaning endeavor. They’re not perfect, and challenges remain, but it would be myopic to ignore them.” —Judge Kevin Newsom
The decision by the U.S. Court of Appeals for the Eleventh Circuit in Snell v. United Specialty Insurance Co. will be cited often for its holdings on policy interpretation and insurance applications under Alabama law. Perhaps the most groundbreaking analysis, likely to have long-term ramifications for insurance coverage litigation, was Judge Kevin Newsom’s concurring opinion addressing the role of AI large language models (LLMs) in policy interpretation. For the first time in a federal appellate decision, a judge openly explored whether ChatGPT, Bard/Gemini and similar AI tools could help courts interpret insurance policy language. His concurrence provides a roadmap for how AI may reshape insurance disputes, and where policyholders must tread carefully.
Below are the key takeaways for policyholders to consider concerning the potential impact of Judge Newsom’s concurring opinion with respect to the use of AI in insurance policy interpretation.
Takeaway #1: Courts Are Beginning to Consider AI as a Tool for Determining “Ordinary Meaning”
Judge Newsom uses his concurrence to discuss whether AI tools could assist courts in identifying the ordinary meaning of undefined policy terms. In Snell, the parties’ claims for coverage involved the definition of a single word—“landscaping.” Judge Newsom writes that traditional interpretation relies on dictionaries, which reflect how editors believe words are used. LLMs, by contrast, train on massive datasets of everyday language and might reflect real-world usage more accurately. LLMs could thus function as “ordinary-meaning engines.”
Judge Newsom’s use of AI tools to interpret the policy term at issue in Snell did not have a practical impact because the majority held that the policyholder had disclaimed coverage for the type of claims at issue in its policy application, regardless of whether the term “landscaping” could be interpreted to include them. But Judge Newsom’s concurrence made clear that responses to AI prompts about the meaning of insurance policy terms could be presented to courts as a policy interpretation tool.
Takeaway #2: Judges Are Consulting AI—Openly or Quietly
Judge Newsom candidly describes experimenting with ChatGPT and Bard/Gemini to see how they defined “landscaping,” noting that the outputs were “more sensible” than expected, surprisingly consistent with his intuitive sense of the word. While he ultimately did not rely on the AI answers to decide the case, he acknowledged that LLMs influenced his thought process. This may mean that, even if AI outputs are not cited in opinions, judges may still consult AI tools informally when interpreting undefined policy terms.
If AI-generated language norms may seep into coverage decisions or subtly influence judicial reasoning, then policyholder counsel must anticipate how AI interprets disputed terms—because judges may already be checking. This creates both uncertainty and opportunity. Policyholders and their counsel must understand how AI tools characterize disputed terms, be prepared to identify weaknesses or inconsistencies in those characterizations, and proactively frame policy language in ways that resist overly narrow or insurer-favored AI interpretations. In short, as AI quietly becomes part of the interpretive landscape, anticipating its influence becomes essential to protecting policyholder rights.
Takeaway #3: Limits, Risks and Flaws of AI Interpretation Pose Special Issues for Insurance Coverage
Well recognized limits and pitfalls of AI could have unintended consequences for policyholders if not identified and challenged. In fact, in another recent concurrence—this one in a criminal case involving the meaning of “physical restraint”—Judge Newsom described prompting an AI model (Claude) for a definition. When he entered the exact same prompt a second time, the model generated a slightly different answer, underscoring just how variable and unpredictable these tools can be.
LLMs continue to routinely exhibit varying degrees of “hallucination,” generating fabricated cases, definitions or factual assertions that could improperly shape arguments or even influence a court’s reasoning if relied upon uncritically. Moreover, because these models are trained primarily on publicly available internet language, their outputs may underrepresent certain industries and linguistic patterns, creating a false sense of comprehensiveness. What may look like a thorough historical account of an ambiguous policy term is only a reflection of the data the model has seen, which may be narrower than expected.
The growing interest in using AI to interpret policy language comes with serious limitations and risks for coverage litigation. LLMs are highly sensitive to how questions are framed, can behave unpredictably when confronted with issues more complex than defining a single term, and may evolve over time as models are retrained, making their outputs difficult to replicate. This raises the specter of an entirely new category of “expert” litigation: Parties may soon feel compelled to hire generative-AI prompt specialists to document their methods, justify their queries and defend AI outputs as sufficiently reliable for judicial consideration. Rather than simplifying interpretation, AI may introduce new layers of complexity and uncertainty into coverage disputes.
Takeaway #4: Policyholders Should Prepare for AI-Infused Insurance Litigation
Policyholders should anticipate a coverage landscape in which AI shapes not just legal arguments but multiple stages of insurance disputes, from claim outcomes to pleadings to discovery to expert retention. As courts and insurers begin consulting AI models to interpret policy language, the use of AI-generated “ordinary meanings” could influence how ambiguities are framed, how exclusions are argued, and how adjusters or judges think about undefined terms.
As an initial matter, companies should consult with experienced coverage counsel in the placement of their policies. Such counsel can identify terms as to which ambiguities may arise and facilitate AI analysis of that term, for use with underwriters or to document understanding and intent for any future coverage disputes.
In the event of a litigated disputed claim, parties may seek discovery about whether insurers used AI tools during underwriting, claims handling or policy drafting, opening the door to disputes over the reliability, transparency and methodology of those tools. Expert witnesses may also be required to explain how AI models operate, what datasets they rely on, why different models produce inconsistent results, and whether any output is sufficiently dependable to inform legal interpretation. Experts may be needed to demonstrate how AI datasets may skew insurer-friendly or fail to capture industry-specific usage, reinforce that established legal principles—such as construing ambiguities in favor of the insured—must control over probabilistic AI predictions, and raise early objections to the improper use of AI-generated definitions in briefs or expert reports.
As AI becomes intertwined with coverage disputes, policyholders who understand its limitations and strategically challenge its use will be best positioned to protect their rights.