Artificial Intelligence (AI) has rapidly advanced in recent years, transforming various industries and impacting our daily lives. In the legal realm, the application of AI is swiftly becoming the subject of an increasing number of lawsuits, some of which have been appealed to the U.S. Supreme Court. In this blog, we will analyze several noteworthy cases and developments and discuss their implications.
Anthropic – Training as Fair Use:
This lawsuit involves a dispute over intellectual property rights related to materials used to train AI technologies. Anthropic PBC, the developer of the “Claude” large language model (LLM), used copyrighted books to train Claude. Anthropic obtained these books in two ways. One way was to buy copies of the books. Anthropic would then tear off the bindings, scan every page, and then store those books in digitized, searchable files. The other way was by downloading pirated copies of “millions” of books from the internet. Anthropic’s goal was to amass a central library of “all the books in the world” to retain “forever.”
Authors of some of the books used for training brought a class action suit in the U.S. District Court for Northern California, alleging infringement. Anthropic claimed that using the books for training constituted fair use and was therefore not infringement. Anthropic moved for summary judgment.
The Court (Alsup, J.) issued what may fairly be called a split decision (Bartz et al. v. Anthropic PBC, No. 3:24-cv-05417 (N.D. Cal. June 23, 2025)). Using the books that were purchased was fair use. The works were transformed: the language was “cleaned” to remove low-value or repeating text, cleaned books were translated into different segments according to the relationships between words, and then compressed, or “memorized,” by the model being trained. No copies of the books were accessible to the public.
The pirated copies presented a different issue, and the court denied summary judgment regarding those pirated books. The piracy of otherwise available copies “is inherently, irredeemably infringing even if the pirated copies are immediately used for the transformative use and immediately discarded”. The court ordered that the plaintiffs may proceed with their infringement claim regarding the pirated works.
This is the first case to consider whether training AI models can be fair use of copyrighted materials. The court has provided clear guidance, especially with regard to the use of pirated materials. Fair use is a fact-intensive inquiry, and it is unrealistic to make an across-the-board prediction as to what courts might hold.
Copyright Office Guidance
In 2023, the U.S. Copyright Office began an examination of the copyright law and policy issues raised by AI. The examination includes the scope of copyright in AI-generated works and the use of copyrighted materials in AI training. The Office is issuing a Report in several Parts analyzing the issues, which will be published as they are completed. Two Parts – Digital Replicas, and Copyrightability – have been issued, while a third – Generative AI Training – was issued in a pre-publication status in May of 2025 (the Copyright Office expects that the Final Version “will be published in the future, without any substantive changes expected in the analysis or conclusions”).
A report from the Copyright Office does not have the binding effect of the law, but such Reports are nonetheless significant. Courts may rely on these Reports when the issues discussed come before them. Perhaps more importantly, the Reports provide guidance to legal professionals and businesses faced with these issues, who are looking for ways to proceed without the expenditure of time and money involved in getting guidance through litigation.
AI-generated Citations Continue to Raise Ethical Concerns:
In an extensively reported case from 2023, two attorneys found themselves in a predicament for their use of AI, specifically, ChatGPT, which generated false citations in their court documents in the Mata v Avianca, Inc., (S.D.N.Y., 1:22-cv-01461) filings. The attorneys were sanctioned by the Court, and were the subjects of significant negative publicity for their actions.
The significant publicity – as well as the sanctions imposed – may have acted as an effective deterrent to attorneys overusing AI for research. One recent case in the Minnesota Tax Court, however, involved an attorney representing a county in property tax litigation. The attorney in Delano Crossing 2016, LLC v. County of Wright, No.: 86-CV-23-2147 (Minn. Tax Ct. May 29, 2025), submitted a brief in support of a motion for summary judgment that included five case citations generated by AI, none of which referred to an actual judicial decision. The attorney claimed that the brief was filed inadvertently. She further attested that, although the case citations in her brief were fake, “the legal contentions in the motion are warranted by existing law as cited [orally during the hearing].”
The court was unconvinced by this reasoning. Rule 11 sanctions were not ordered, but a referral was made to the Minnesota Board of Lawyers’ Professional Responsibility, alleging that “the submission of an AI-generated brief, apparently unreviewed, as evidenced by inclusion of entirely fake case citations, reasonably raises questions as to a lawyer’s honesty, trustworthiness, and/or fitness as a lawyer.”
The case underscores the importance of checking research before making a court submission, particularly when AI is involved. Although AI can be a valuable research tool, it should not be the last step in performing research. Careful review is essential.
Tennessee Fights Deepfakes
Tennessee has, for many years, taken a strong interest in protecting the rights of celebrities to their own likenesses, including their voices as well as their appearance. This is understandable, given the location of two centers of American popular music – Memphis and Nashville – in the state. The protection continues even after the death of the celebrity.
In 2024, a law went into effect in Tennessee that considers a new facet of protecting celebrity likenesses. The “Ensuring Likeness, Voice, and Image Security [ELVIS] Act of 2024,” Tenn. Code Ann. § 47-25-1101, et seq., prohibits using or distributing “an algorithm, software, tool, or other technology, service, or device, the primary purpose or function of [which] is the production of a particular, identifiable individual’s photograph, voice, or likeness,” knowing that this use was not authorized by the individual or their estate. Tenn. Code Ann. § 47-25-1105. The inclusion of a person’s voice as protected is especially significant. It is relatively easy to manipulate a recording to the extent that a listener might think they were listening to a different performer. The new law prohibits that manipulation. The law is relatively new, so there has not been significant litigation about artificially generated imitations. If or when it does happen, it may be expected that the creator of the digital imitation may claim that the AI-generated imitation is a parody or commentary on the original work. While such a defense is allowed by statute, it will remain to be seen whether such a defense will be viable for a truly convincing AI deepfake.