Judge dismisses most of Sarah Silverman’s copyright infringement lawsuit against OpenAI

It’s been a pretty good week over at OpenAI, so far. Despite a small protest outside its headquarters last night, the company shipped an update to ChatGPT to give it persistent memory, and yesterday, it saw most of the class-action copyright infringement lawsuit filed against it by comedian Sarah Silverman over allegedly copying her 2010 book, The Bedwetter, as part of its massive data scraping effort to train its AI models dismissed by the judge presiding over the case.

Quick background on the case

In fact, in the case of Silverman et al v. OpenAI, Inc. in the U.S. District Court of Northern California (which oversees San Francisco and much of Silicon Valley), Judge Araceli Martínez-Olguín ruled against four of the six counts in the lawsuit originally filed in July 2023 by Silverman and her co-plaintiffs, authors Richard Kadrey and Christopher Golden, who all accused OpenAI of violating their copyrights by training its AI models GPT-3.5 and GPT-4 on their respective books without consent, to power its ChatGPT consumer facing application.

OpenAI had sought to dismiss five of the plaintiffs’ original six counts of infringement, and it won most of them, though Judge Martínez-Olguín invited the plaintiffs to amend their complaint and re-file by March 13 of this year — a month from today.

What the judge ruled

The only count the judge is allowing to proceed against OpenAI is one that states by training its commercial AI models on their books, OpenAI committed “unfair” business practices, a violation of California state law (the Unfair Competition Law, or UCL).

As the judge wrote in her order:

“Assuming the truth of Plaintiffs’ allegations – that Defendants used Plaintiffs’ copyrighted
works to train their language models for commercial profit – the Court concludes that Defendants’
conduct may constitute an unfair practice. Therefore, this portion of the UCL claim may proceed.“

That’s definitely not what OpenAI was hoping, but still, the decision overall is largely a victory for the fast growing AI company, especially given the counts that were dismissed.

Wider implications

Indeed, with the caveat that I’m a written journalist and customer of OpenAI’s via my ChatGPT Plus personal subscription — and have no formal legal training or expertise — it seems like the judge’s arguments in this case bode well for generative AI companies overall as they face down lawsuits from creatives and rights holders’ who contest them training on their copyrighted works without express permission or consent, let alone compensation. That is, if the courts rule similarly in other jurisdictions.

Ultimately, the judge in this case observed that the lawyers representing Silverman and her co-plaintiffs’ did not present enough or any evidence that ChatGPT was copying their books wholesale, or even significant portions of them, in its responses to users. The judge and by extension, court, did conclude that OpenAI copied the books for training purposes, on the back-end, but did not reproduce them on the front-end for paying customers, meaning it did not violate copyright.

Copying copyrighted works to produce summaries isn’t infringement

As the latest decision document states: “OpenAI copied Plaintiffs’ copyrighted books and used them in its training dataset. When prompted to summarize books written by each of the Plaintiffs, ChatGPT generated accurate summaries of the books’ content and themes.”

However, as the judge explains: “Distinctly, Plaintiffs here have not alleged that the ChatGPT outputs contain direct copies of the copyrighted books...Plaintiffs fail to explain what the outputs entail or allege that any particular output is substantially similar – or similar at all – to their books. Accordingly, the Court dismisses the vicarious copyright infringement claim with leave to amend.“

In other words: just because OpenAI ingested the entire contents of the books for training purposes and ChatGPT is capable of accurately summarizing them, does not mean those summaries or other responses it returns about the books are inherently infringing. The plaintiffs’ lawyers didn’t show enough examples of direct copying and infringement taking place in the form of ChatGPT’s responses.

The lawyers representing Silverman and her co-plaintiffs also argued that OpenAI violated copyright by removing “copyright management information” when copying the books for its AI training — after all, this information doesn’t reappear in ChatGPT’s summaries. But as the judge ruled, “Plaintiffs do not plausibly allege that OpenAI intentionally removed CMI during the training process or intended to conceal or induce infringement.”

But did OpenAI violate the Digital Millennium Copyright Act by creating “derivative works” of Silverman’s, Kadrey’s and Golden’s books in the form of “ChatGPT outputs” without the proper CMI attached to it? Again, the judge says not so much, noting the plaintiffs’ lawyers “have alleged that ‘every output from the OpenAI Language Models is an infringing derivative work’ without providing any indication as to what such outputs entail – i.e., whether they are the copyrighted books or copies of the books. That is insufficient to support this cause of action under the DMCA.“

Furthermore, the judge noted that Silverman and co.’s lawyers “have not alleged that OpenAI unjustly obtained benefits from Plaintiffs’ copyrighted works through fraud, mistake, coercion,” and “do not explain how merely possessing their books creates a special relationship,” wherein OpenAI would have been contractually obligated to maintain and control the information from their books in its possession in a certain way.

Despite all this, the case is not fully resolved, and it will really come down to how the lawyers for Silverman and her co-plaintiffs are able to amend their claims to see whether or not it can proceed to a full trial. Until then, here’s the latest decision embedded for you to review: