[EDRM Editor’s Note: The opinions and positions are those of Marris Hoffee and Michael Berman.]
Sometimes, sophisticated analysis is needed to detect a misuse of A.I. Sometimes, it isn’t.
In Dillard v. CBS Studios, Inc., 2026 WL 790882 (C.D. Cal. Mar. 16, 2026), the plaintiff brought a copyright infringement action that quickly unraveled under scrutiny.
The court noted: “Plaintiff’s submissions in this case may have been prepared by using generative artificial intelligence,” noting that the filings “misrepresented the Works at issue here” and “mischaracterized Defendant’s conduct and the prior orders issued in this case.”
[S]ystematic factual misrepresentations are indicia of generative artificial intelligence…
Dillard v. CBS Studios, Inc., 2026 WL 790882 (C.D. Cal. Mar. 16, 2026).
The court emphasized that such defects reflected a broader breakdown in reliability. Citing precedent, the court explained that “systematic factual misrepresentations are indicia of generative artificial intelligence…” It added: “Plaintiff’s unauthorized supplemental briefing contains images that appear to have been created by generative artificial intelligence.” In support, the court wrote:
The images in this brief contain nonsensical spellings that are characteristic of images generated by artificial intelligence. See Dkt. 64 at 12 (“Twombly/idqal + Three Bays”); id. (“Access need not be proven beyond doubt–only alausibly.”); id. at 11 (“Comedie Freeze”); id. at 11 (“cafeteira tone”); id. at 11 (“Misúrstranding”); see also Amanda Silberling, Why Is AI So Bad At Spelling? Because Image Generators Aren’t Actually Reading Text, TechCrunch (Mar. 21, 2024), https://techcrunch.com/2024/03/21/why-is-ai-so-bad-at-spelling/ (discussing similar nonsensical spellings in images generated by artificial intelligence).
Id. at n. 12 (emphasis added).
Significantly, the court also identified “hallucinated” cases – citations to false or nonexistent legal authority.
The court refused to grant leave to amend the Complaint. This was not a curable pleading defect. Rather, it damaged the integrity of the litigation itself. The court concluded that the “unsupported theories and legal contentions advanced by Plaintiff, which appear to be the result of his use of generative artificial intelligence, further confirm that granting leave to amend will allow Plaintiff to raise additional unsupported allegations.” In other words, the problem was not merely insufficient pleading – it was the demonstrated willingness to rely on fabricated authority and misrepresentations.
Accordingly, Count I was dismissed without leave to amend, reflecting that the court need not, and will not, permit parties to proceed where their pleadings are grounded in nonexistent, AI generated authority. Other counts were dismissed on separate substantive grounds not relevant here.
Assisted by GAI and LLM Technologies per EDRM’s GAI and LLM Policy.
