Anthropic Settlement with Authors Could Set an AI Industry Precedent

“It appears that under the current law, so long as you have lawfully procured a copy of a work, it is fair use to use that as training material in a neural network on the basis that the actual use is considered transformative fair use.” – Randy McCarthy, Hall Estill

AnthropicAccording to a Consent Motion filed Tuesday, August 26, with the U.S. Court of Appeals for the Ninth Circuit, Anthropic has requested that its petition appealing a district court ruling on class certification and its Emergency Motion to stay district court proceedings pending disposition of that petition both be held in abeyance.

The request is based on the execution by both Anthropic and plaintiffs Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson of “a binding term sheet intended to memorialize the terms of a proposed class settlement.”

Bartz et. al., who are journalists and book authors, filed the lawsuit in August 2024 alleging that Anthropic’s widespread copyright infringement involves “hundreds of thousands of copyrighted books” from unauthorized sources used to train Anthropic’s Claude AI chatbot. Judge William Alsup denied Anthropic’s motion to stay proceedings pending appeal earlier this month, following Anthropic’s request for permission to file an interlocutory appeal of Alsup’s June order on fair use.

The order on fair use granted summary judgment for Anthropic that its use of the works at issue for training and its scanning of certain works from print-to-digital format were fair, but denied summary judgment for Anthropic that certain pirated library copies of the relevant works must be treated as training copies and ordered a trial with respect to the pirated copies to determine damages, including potentially for willfulness.

Anthropic asked Alsup to certify an order for interlocutory appeal or, alternatively, leave to file a motion for reconsideration of the Order, citing a ruling issued just days after Alsup’s order in which Judge Vince Chhabria said  a number of well-known authors failed to successfully argue that the market for their works was significantly harmed by Meta’s use of the works to train its generative AI tool, Llama. In view of this split, said Anthropic in its motion, “[t]his Court should obtain guidance from the Ninth Circuit on the issue now instead of holding a trial that may need to be redone under a different legal framework—or may not be necessary at all.”

However, Anthropic’s decision to settle is not a surprise in light of the way the order panned out, according to Randy McCarthy of Hall Estill. In comments sent to IPWatchdog, McCarthy said that “once the question became one of potential liability for having used pirated materials, which is fairly straightforward,” settlement became likely.

“It appears that under the current law, so long as you have lawfully procured a copy of a work, it is fair use to use that as training material in a neural network on the basis that the actual use is considered transformative fair use,” McCarthy added.

The settlement leaves open the question of whether copyright owners can deny companies permission to use their works for training purposes, however, McCarthy noted. End user license agreements (EULA) are becoming one way to address this, and the settlement itself could set a business precedent for AI companies going forward, some commentators have said.

McCarthy pointed to the pending lawsuit brought by Disney and Universal Studios against Midjourney as one in which these issues could be hashed out further. There, Disney and universal alleged that Midjourney is a “bottomless pit of plagiarism” and that it could have stopped the infringement and copying of their copyrighted works at any time—either by controlling the data used to train, by controlling the prompts users input, or via technological protection measures—but chose not to and failed to respond to letters informing them of the infringement prior to the lawsuit.

The Disney/ Universal lawsuit also points to statements made by Midjourney CEO David Holz indicating that he would be open to an opt out model so that copyright owners could control whether their content is used for training, but that the company doesn’t “have a process for that yet.”

Attorneys for the authors in the case against Anthropic have said in statements to the press that the details of the settlement will be revealed in the coming weeks.

Image Source: Deposit Photos
Author: rafapress
Image ID: 649477252

Share

Warning & Disclaimer: The pages, articles and comments on IPWatchdog.com do not constitute legal advice, nor do they create any attorney-client relationship. The articles published express the personal opinion and views of the author as of the time of publication and should not be attributed to the author’s employer, clients or the sponsors of IPWatchdog.com.

Join the Discussion

One comment so far.

  • [Avatar for Anon]
    Anon
    August 28, 2025 09:56 am

    I have seen this view by more than Mr. McCarthy:

    It appears that under the current law, so long as you have lawfully procured a copy of a work, it is fair use to use that as training material in a neural network on the basis that the actual use is considered transformative fair use,

    This would be error, as it conflates the separate acts and draws a legal conclusion not reached expressly due to the settlement removing the question from the court’s purview.

    This also:

    The settlement leaves open the question of whether copyright owners can deny companies permission to use their works for training purposes

    is unsupported error.

    Fair Use simply does not require permission. There is no question then that any use that is Fair Use provides copyright owners any legal action that is simply not within their granted rights.

Varsity Sponsors

Industry Events

PIUG 2026 Joint Annual and Biotechnology Conference
May 19 @ 8:00 am - May 21 @ 5:00 pm EDT
Certified Patent Valuation Analyst Training
May 28 @ 9:00 am - May 29 @ 5:00 pm EDT
2026 WIPO-U.S. Summer School on Intellectual Property
June 1 @ 9:00 am - June 12 @ 1:45 pm EDT

From IPWatchdog