Artificial Intelligence in Litigation: A Tool, Not a Witness

Artificial intelligence is no longer a future issue in litigation. It is already embedded — sometimes quietly, sometimes explicitly — in document review, damages analysis, expert modeling, and case assessment. As its use expands, courts and practitioners are increasingly focused on a practical question: how artificial intelligence fits within existing evidentiary, ethical, and professional frameworks.

So far, the answer has been evolutionary rather than disruptive. Courts have shown little inclination to treat AI as something that displaces settled legal standards. Instead, judges appear to be applying familiar principles — reliability, relevance, transparency, and professional judgment — to AI-assisted work, including expert testimony.

AI as an Analytic Tool, Not an Opinion Source

Experts have relied on technology for decades. Statistical software, regression tools, valuation models, and large-scale databases are routine components of forensic accounting and damages analyses. Artificial intelligence belongs in that lineage, but with important distinctions.

Unlike traditional tools, many AI systems — particularly generative models — can produce outputs that appear authoritative without revealing the reasoning behind them. That characteristic may be useful for preliminary analysis or information gathering, but it creates risk when such output is used to support opinions offered under oath.

Courts evaluating expert testimony under Federal Rule of Evidence 702 focus on whether opinions are the product of reliable principles and methods, reliably applied to the facts of the case. That inquiry derives directly from Daubert v. Merrell Dow Pharmaceuticals, Inc., which assigns the trial judge a gatekeeping role to ensure that expert testimony rests on a reliable foundation and will assist the trier of fact.^[i] Artificial intelligence does not alter that analysis. If anything, its use may invite closer scrutiny because opacity, training-data bias, and hallucinated output raise questions about testability, validation, and error rates—factors long associated with admissibility.

A Recent Warning From the Courts

The risks of uncritical AI use are no longer hypothetical. In Kohls et al. v. Ellison et al., a 2025 decision from the U.S. District Court for the District of Minnesota, the court addressed expert declarations submitted in connection with a challenge to Minnesota’s “deepfake” statute.^[ii] One expert — retained to explain the dangers of AI-driven misinformation — acknowledged that his declaration contained citations to non-existent academic articles generated by an AI tool.

The court denied a request to rehabilitate the declaration through amendment, noting that a sworn submission supported by fabricated citations undermined reliability and credibility. While the case arose in a constitutional challenge rather than a traditional damages dispute, the lesson translates directly: courts expect experts to independently verify all sources and materials underlying their opinions, regardless of whether AI assisted in drafting, research, or analysis.

Notably, the court did not announce a rule prohibiting AI use. Instead, it reinforced a long-standing principle—experts who fail to exercise independent judgment and verification expose themselves to serious admissibility and credibility risk.

Disclosure: Context Matters

Whether and how an expert should disclose the use of artificial intelligence is an evolving question. There is no universal rule, and absolutism in either direction is unwarranted. That said, disclosure should often be considered, particularly where AI meaningfully influenced the analysis, methodology, or presentation of opinions.

Expert reports already require disclosure of facts, assumptions, and methods. When AI tools materially affect those elements, acknowledging their role may enhance credibility rather than diminish it—provided the expert can clearly explain how the tool was used, what safeguards were applied, and what aspects of the work were independently verified.

Disclosure does not require identifying every prompt, parameter, or proprietary workflow. Courts generally care less about the brand or label of the tool than about whether the expert understands it, supervised its use, and can explain and defend the resulting opinions.

Human Judgment Remains Central

Across judicial opinions, ethics guidance, and professional literature, one theme is consistent: human verification matters. Experts may use AI to surface patterns, test assumptions, or manage large datasets, but they remain responsible for every conclusion offered.

Academic and professional literature in forensic accounting consistently treats AI as a complement to professional expertise and skepticism—not as a substitute.^[iii] Courts evaluating admissibility under Rule 702 similarly focus on whether a method was applied thoughtfully and connected to the specific facts of the case. AI-assisted output accepted uncritically, without testing or corroboration, presents vulnerabilities that opposing counsel — and judges — are likely to exploit.

Privilege, Confidentiality, and the Record Created

AI also raises practical litigation risks beyond admissibility. Courts are increasingly treating AI prompts and outputs as ordinary electronically stored information. Use of public or open AI platforms may also implicate confidentiality and privilege, particularly where platform terms permit retention or reuse of user inputs.

Recent federal decisions addressing privilege claims tied to generative AI use emphasize that traditional, technology-neutral principles apply: confidentiality depends on how the communication was made and whether reasonable safeguards were in place—not on the novelty of the tool itself.^[iv] Experts engaged in litigation matters should therefore consider, at the outset, where and how AI tools are used, especially when handling sensitive or case-specific information.

No Absolutes—But Clear Expectations

Artificial intelligence will continue to play a role in litigation and forensic work. Courts are not resistant to that reality. What they resist is the notion that AI changes the expert’s fundamental obligations.

Experts remain expected to understand their tools, verify their work, disclose material influences where appropriate, and exercise independent judgment. Artificial intelligence may enhance efficiency and insight, but it does not testify. The expert does.

At CBIZ Forensic Consulting Group, LLC, we view AI as one of many tools available to experienced professionals. Used thoughtfully, and transparently when appropriate, it can strengthen expert analysis. Used carelessly, it can undermine credibility faster than almost any cross-examination.

Note: Portions of this article were originally drafted using AI; research/endnotes were initially generated by AI research and checked by human experts.

Reprinted with permission from the June 8, 2026, edition of the “Legal Intelligencer” © 2026 ALM Global Properties, LLC. All rights reserved. Further duplication without permission is prohibited, contact 877-256-2472 or [email protected].