Bloomberg must face a proposed copyright infringement class action led by former Arkansas Gov. Mike Huckabee alleging the media company used e-books to train its large language model, after a New York federal judge said Monday she can't determine whether the fair use defense applied without "a robust factual record."

In a seven-page order, U.S. District Judge Margaret M. Garnett of the Southern District of New York denied a motion to dismiss filed last year by Bloomberg LP and Bloomberg Finance LP against a lawsuit filed by a group of authors, including Huckabee, claiming the defendants trained their AI tools on datasets that pirated approximately 183,000 e-books.

Bloomberg contended the authors' copyright claim is barred under the doctrine of fair use and that there's no plausible allegation that its LLM, BloombergGPT, has in any way replaced or displaced their books in the literature market. Bloomberg further argued the plaintiffs' first amended complaint alleged that the defendant, as a provider of financial news and analytics, used their works to teach a generative AI model as part of a noncommercial initiative.

On Monday, Judge Garnett said Bloomberg hasn't established it's entitled to the fair use defense at this juncture, largely because she needs a "robust factual record" to assess whether that defense applied.

There are four factors in the fair use analysis: the purpose and character of the use, nature of the copyrighted work, the amount and substantiality of the part used in relation to the copyrighted work as a whole, and the impact of the use upon the copyrighted work's potential market or value, the ruling said.

Judge Garnett rattled off a series of outstanding issues she'd have to address conducting the fair use analysis in the instant dispute, including the fourth fair use factor.

"In this case, analyzing this factor would conceivably require the court to address at least the following questions: Will widespread copying of literary works to train LLM AI programs result in a deluge of AI-produced literary works, adversely affecting the market for the Works?" the judge wrote. "Because LLM AI programs have the potential to yield massive profits but require significant amounts of text to train, is there a derivative market for literary works used to train AI?"

The judge continued: "Just these few examples illustrate the futility of conducting a fair use analysis on the face of a complaint in this complex and evolving area of both law and technology, without a factual record," the judge said.

In a footnote, the judge referenced other factors under the fair use analysis that pose similar issues, like deciding whether the challenged use was commercial in nature. Again, there needs to be a sufficient factual record to make that finding, the order said.

"And, drawing all inferences in plaintiffs' favor, the amended complaint suggests that Bloomberg's development of BloombergGPT was for a commercial purpose," the footnote said. "Bloomberg is a finance news company. It allegedly built BloombergGPT as the world's first LLM AI program for 'finance.' The plausible inference is that BloombergGPT will assist Bloomberg in conducting financial reporting, the core of Bloomberg's line of business, or developing new information products to sell to its financial industry customers."

Monday's order is the latest development in the suit that was initially filed in October 2023, which also named Meta and Microsoft as defendants.

Microsoft and Meta agreed with the plaintiffs to sever and transfer all claims to the Northern District of California, where separate suits involving comedian Sarah Silverman and author Michael Chabon have been consolidated.

Bloomberg moved to dismiss the suit, which was amended in January 2024 to leave the media company as the only defendant in New York, and its one claim of direct copyright infringement. Bloomberg argued the plaintiffs don't allege that BloombergGPT had been publicly released, that it had been used in any commercial capacity or that it's even been used by anyone outside the company.

Nor do the authors allege which of their copyrights works were purportedly used to train Bloomberg's LLM, the defendants said. Bloomberg further argued that the first amended complaint alleged Bloomberg — as a provider of financial news and analytics — used the plaintiffs' works to teach a generative AI model as part of a noncommercial research initiative.

On Monday, Judge Garnett also rejected Bloomberg's argument that the suit failed to state a claim since it didn't precisely identify which works were within datasets used for training, noting that the suit does plausibly allege a copyright infringement claim.

The plaintiffs' complaint says the Books3 dataset included the copyrighted works at issue, and referenced a research paper that Bloomberg published, which said Bloomberg used the data contained in Books3 to train BloombergGPT, the judge pointed out. The complaint also alleges that the use of their works happened without the plaintiffs' permission or authorization.

"This provides fair notice and a plausible allegation of the conduct plaintiff contend was copyright infringement," Judge Garnett said.

Representatives for the Bloomberg entities did not immediately respond to inquiries seeking comment Tuesday.

Nada Djordjevic of DiCello Levitt, one of the attorneys for the proposed class, issued the following statement Tuesday afternoon: "We're happy with Judge Garnett's well-reasoned decision to deny Defendants' motion to dismiss in all respects, and we look forward to successfully prosecuting our important copyright infringement and other claims — on behalf of Ambassador Huckabee and the other class members — through trial."

The plaintiffs are represented by Adam J. Levitt, Greg G. Gutzler, Amy E. Keller, James A. Ulwick and Nada Djordjevic of DiCello Levitt LLP, Seth Haines and Timothy Hutchinson of RMP LLP, and Scott Poynter of Poynter Law Group.

The Bloomberg entities are represented by Nicole M. Jantzi, Paul M. Schoenhard and Amir R. Ghavi of Fried Frank Harris Shriver & Jacobson LLP.

The case is Mike Huckabee et al. v. Meta Platforms Inc. et al., case number 1:23-cv-09152, in the U.S. District Court for the Southern District of New York.