The Courts Have Spoken: AI Isn't the Problem — Hiding It Is

If you're a forensic evaluator who has hesitated to use AI tools in any part of your professional workflow, you're not alone. The uncertainty is understandable. Headlines about sanctioned experts and excluded testimony make it easy to conclude that AI and the courtroom don't mix.

But that conclusion is wrong. The court rulings tell a far more nuanced — and more encouraging — story than most forensic professionals realize.

What the Cases Actually Say

In April 2025, the Eastern District of New York decided Ferlito v. Harbor Freight Tools. An expert witness — a consultant with extensive practical experience in engineering and power tools, though without formal engineering degrees — had used ChatGPT after completing his report to confirm his findings, which were based on his hands-on professional experience. Opposing counsel moved to exclude the testimony, arguing in part that the AI use undermined reliability.

The court refused. The critical distinction was straightforward: the expert relied on his own professional experience and judgment to form his opinions first, then used AI only as a confirmatory check. The AI did not generate the opinions. It verified them.

This ruling didn't just allow AI use — it articulated a framework that validates exactly how most conscientious professionals would want to use AI: as a quality-control layer applied after you've done your clinical work.

Where It Goes Wrong

The cases where courts sanctioned experts tell an equally clear story — and the lesson is not "don't use AI." The lesson is "don't be careless with AI."

In Kohls v. Ellison (D. Minn., January 2025), Stanford professor Jeff Hancock — ironically an expert on AI-generated deepfakes and misinformation — submitted a declaration that cited two nonexistent academic articles and misattributed a third. The fabricated citations had been generated by ChatGPT-4o. The court struck the entire declaration, finding that the fictitious citations "shatter[ed] his credibility" and noting that the expert had failed to apply the same verification standards he used in his academic work. The court stated that the expert had "abdicate[d] his independent judgment and critical thinking skills in favor of ready-made, AI-generated answers."

In the Concord Music Group case (N.D. Cal., May 2025), a hallucinated citation — a nonexistent article title with inaccurate authors who had never actually collaborated — led the court to strike the affected paragraph and note that the incident undermined the credibility of the expert's entire declaration. Notably, the error originated not with the expert herself but with the legal team: a Latham & Watkins associate had used Claude.ai to format an existing citation, and the AI returned a fictitious article title and wrong authors while preserving the correct link and publication details. The underlying article was real, which the court acknowledged made the situation less grave than it first appeared — but the failure of the manual cite-check process remained a serious concern.

The pattern across these cases is clear. Courts are not hostile to AI. They are hostile to unverified AI output presented as the expert's own work.

The Framework That's Emerging

Across these rulings and the positions of organizations like the American Board of Professional Psychology — which has published guidance on AI in forensic practice and convened an AI Task Force — a general framework is emerging around what constitutes defensible versus indefensible AI use.

Defensible AI use looks like this: the expert forms opinions independently first. AI is used as a quality-control or verification layer. The expert can explain every finding without reference to AI. All AI-generated content is independently verified. AI use is documented and disclosed when asked.

Indefensible AI use looks like this: AI generates the opinions or conclusions. AI serves as the primary analytical engine. The expert cannot reproduce or defend findings independently. AI output is included without verification. AI use is concealed or minimized when discovered.

Notice that the dividing line is not whether you used AI. It's whether you maintained your role as the independent expert and used AI to support — rather than replace — your professional judgment.

The Disclosure Reality

This brings up the question many evaluators are quietly wrestling with: if I use AI, do I have to tell anyone?

The honest answer is that the landscape is evolving rapidly, and the safest posture is transparency. Texas, for example, enacted Senate Bill 1188 in 2025, which requires healthcare practitioners to disclose any use of AI in diagnosis or the course of treatment — with the disclosure provided in clear, conspicuous, plain language. Proposed Federal Rule of Evidence 707 would require AI-generated evidence to meet reliability standards analogous to those established in Daubert v. Merrell Dow Pharmaceuticals. Courts are also beginning to issue standing orders requiring AI disclosure in filings and submissions.

But here's the reframe that matters: disclosure is a strength, not a vulnerability. The courts in Ferlito, Kohls, and Concord Music all showed far greater concern about concealment and unverified use than about AI use itself. An expert who proactively describes their AI-assisted quality assurance process demonstrates rigor. An expert who gets caught having used AI without mentioning it faces a credibility crisis that has nothing to do with the quality of their clinical work.

When asked "Did you use AI in preparing your report?" the strongest answer isn't a defensive one. It's something like: "I wrote my report based on my clinical evaluation and professional judgment. After completing it, I used a HIPAA-compliant methodology review tool to check for potential vulnerabilities — similar to how I might ask a colleague to review my work. My clinical opinions, methodology, and conclusions are entirely my own."

That answer demonstrates preparation, not weakness. And it's the answer the courts are rewarding.

What This Means for Your Practice

The window of ambiguity around AI use in forensic practice is closing. Rules are being written — through case law, through proposed federal rules, through state legislation — and they are converging on a single principle: transparent, supplementary, verified AI use is defensible; undisclosed AI ghostwriting is not.

Forensic professionals who adopt AI tools thoughtfully now, with proper documentation and disclosure practices, will be positioned on the right side of that line. Those who either avoid AI entirely or use general-purpose tools without safeguards will find themselves increasingly disadvantaged — either underprepared relative to opposing counsel, or exposed to the kind of credibility attacks that ended careers in Kohls and Concord Music.

The courts have spoken. The question is whether you're listening.

ForensicShield is designed around the exact framework courts are rewarding: you do your clinical work first, then the tool reviews your completed report for vulnerabilities. Your opinions stay yours. The AI supports your preparation — nothing more. Learn how it works →

References

American Board of Professional Psychology. (2025). Guidance on the use of artificial intelligence in professional psychology.

Buchanan, A., Norko, M., Baranoski, M., & Zonana, H. (2024). Balance and change in forensic psychiatry. Journal of the American Academy of Psychiatry and the Law, 51(3), 345–352. https://doi.org/10.29158/JAAPL.230045-23

Concord Music Group, Inc. v. Anthropic PBC, No. 5:24-cv-03811 (N.D. Cal. May 2025).

Daubert v. Merrell Dow Pharmaceuticals, Inc., 509 U.S. 579 (1993).

Ferlito v. Harbor Freight Tools USA, Inc., No. 20-cv-05615 (E.D.N.Y. Apr. 2025).

Kohls v. Ellison, No. 24-cv-03754 (D. Minn. Jan. 2025).

Proposed Federal Rule of Evidence 707. (2025).

Texas Senate Bill 1188, 89th Leg., Reg. Sess. (2025).

Note: Some cases cited are recent federal district court decisions not yet published in official reporters; citations reflect currently available court records.