Ways to Sue a Generative A.I. Company and (Possibly) Win
A few legal theories that have survived motions to dismiss and could possible allow plaintiff to win against generative artificial intelligence (AI) companies are claims of trademark dilution (federal and state), false endorsement and trade dress (Lanham Act).
State law claims against generative AI companies are generally preempted by copyright right law. see New York Times Co. v. Microsoft Corp., 777 F. Supp. 3d 283, 320 Dist. Court, SD New York 2025. It is in the interest of generative AI companies to get state law claims pre-empted into copyright claims because copyright law has a broad and fact specific fair use defense. see 17 U.S.C. § 107. While trademark dilution, false endorsement, and trade dress have their own defenses (including fair use), they are far less expansive.
In New York Times Co. v. Microsoft Corp., 777 F. Supp. 3d 283, 320 Dist. Court, SD New York 2025, the Court denied Open AI’s motion to dismiss the claims based on trademark dilution (based on federal and state law claims) finding the New York Times pleading that Open AI was using its trademark without authorization on lower-quality and inaccurate writing. See Id. at 323-326.
Even though generative AI models would not use a visual artist in the same manner, a visual artist could win on a motion to dismiss based on a claim of trademark dilution if their art had a watermark (or any other signature) that the large language model reproduced (even if it was a modified version of it). Getty is making a similar claim in their case against Stability AI. see Complaint at 38, 40 Getty Images (US), Inc. v. Stability AI, Ltd., No. 3:25-cv-06891 (N.D. Cal. Aug. 14, 2025).
In Andersen v. Stability AI Ltd., 744 F. Supp. 3d 956 (N.D. Cal. 2024), the Court denied Open AI’s motion to dismiss the claims based false endorsement and vicarious trade dress infringement. Id. at 977-981. While the false endorsement claim arose from a situation not likely to be repeated (Stability advertised that the model could replicate artists style and specifically referred to them by name), the claim for trade dress is something that could be replicated. Generally trade dress are recognized as "words, symbols, collections of colors and designs, or advertising materials or techniques" that the purchasing public has come to associate with a single source. see Romm Art Creations Ltd. v. Simcha Int'l, Inc., 786 F. Supp. 1126, 1134 (E.D.N.Y. 1992). In Andersen, the Court weighted Stability AI’s references to the artists names as contributing to a finding that the purchasing public has come to associate with a single source, which is unlikely to be repeated. see Andersen, 744 F. Supp. 3d at 977-981. However, there have been other cases where visual artists have been successful in asserting trade dress infringement. see Romm Art Creations Ltd. v. Simcha Int'l, Inc., 786 F. Supp. 1126, 1134 (E.D.N.Y. 1992). One of the hurdles in a claim for trade dress infringement is whether the elements have acquired secondary meaning. However, courts have found that intentional copying of trade dress creates a triable for of secondary meaning. see P&P Imports LLC v. Johnson Enters., LLC, 46 F.4th 953 (9th Cir. 2022). Since the LLMS intentionally copy everything that is available on the internet, this would likely not be an issue.
How can visual artists give themselves the greatest chance in state law or Lanham Act claims? Having a distinctive (but unobtrusive) watermark or signature on anything you post online would likely give artists a fighting chance in court against generative AI under trademark, trade dress infringement, or false endorsement.