Are Tech's Generative AI "Fair Use" Dominoes Starting to Fall?
Recent Court Developments Don't Bode Well for "Fair Use" Defenses for AI Training Without Consent & Compensation to Rights-holders
It’s Monday morning - brAIn dump time! First, it’s this week’s “mAIn event” (my feature story about Tech’s increasingly precarious reliance on “fair use” defenses). Next, it’s your GenAI “video of the week.” Finally, the “AI Litigation Tracker” (updates the key generative AI infringement cases by the law firm McKool Smith).
I. The mAIn event - Big Tech’s “Fair Use” Dominoes Are Starting to Fall
Background/Context.
We all know the central issue that permeates all things related to generative AI and the media and entertainment industry — i.e., whether GenAI developers (the usual suspects) must license the content they need to train their AI models. Copyright owners (media companies, publishers, music labels, Artists) say “yes, absolutely — we must first give consent, and then be compensated.” Big Tech GenAI developers, on the other hand, say “absolutely not — it’s all ‘fair use.’” That’s what all the litigation (highlighted in the “AI Litigation Tracker” below) is all about.
No courts have yet definitively ruled on this “fair use” issue in the context of this GenAI-focused copyright litigation. But you know where I stand — firmly on the side of the creative community and rights-holders on this fundamental and, in many ways, existential issue. It’s an outright “taking” — both ethically, and as a matter of copyright law (I’ve written full legal analyses like this one in earlier newsletters).
My conclusion is not based in fear about this new transformational technology, or an effort to slow its progress (impossible to do anyhow). I ran several pioneering tech-forward media companies after all. It’s simply about doing new technology “right.” The question of doing things “right” in the context of GenAI is, to me, a simple one — if developers need content to train their AI models to make them useful (which they do — endless amounts of it!), then they must pay for that necessary ingredient and fairly share in the wealth that the potent combination of content and tech makes.
Rights-Holders Have 3 Options, Faced with This Re-AI-ity.
So what are the options for a media and entertainment industry faced with unabated, unlicensed, and largely unapologetic scraping of its copyrighted works? Rights-holders have three: (1) do nothing (just let it happen, which benefits the GenAI developers 100%); (2) negotiate licensing agreements with GenAI developers (the best possible course of action, and one which is increasingly happening — although still very much in its nascent stages and not yet public when it comes to video content); or (3) litigate the issue of “fair use,” tech’s fundamental copyright defense that, again, no U.S. court has conclusively addressed.
Option 3’s central “fair use” issue is a high stakes gamble for both sides of this Hollywood/Silicon Valley divide, although I certainly believe that rights-holders have the upper hand.
And that hand is getting stronger and stronger.
(1) “Fair Use” Is Just Different In This Generative AI Context.
First, in August, Judge William Orrick, the federal judge presiding over one of the most important GenAI “training” infringement cases (Northern District of California), rejected the attempt by generative art developer Stability AI (and other defendants) to dismiss claims of infringement on “fair use” grounds as a matter of law (i.e., the “motion to dismiss” phase of litigation). Orrick, in fact, opened the door to deciding “fair use” himself pre-trial on what is known as “summary judgement” (you can read the Judge’s entire written opinion here; I did).
Importantly, in his ruling, Judge Orrick — who had not previously seemed to have been overly sympathetic to the rights-holding plaintiffs — also expressly noted that infringement claims made against GenAI defendants are utterly unique in the context of copyright law — and that “run of the mill” substantial similarity copyright cases are “unhelpful in this case where the copyrighted works themselves are alleged to have not only been used to train the AI models but also invoked in their operation.” That means that the tech world’s reliance on earlier content/tech copyright cases — including the renowned “Google Books” case that I discuss in Section (3) below — may be entirely misplaced.
None of this means that the rights-holding plaintiffs here will ultimately prevail, but it certainly doesn’t hurt. Now the parties move on to the evidentiary phase on the issue of “fair use.” More on that below in point (5). (NOTE: I previously wrote an entire analysis of Judge Orrick’s ruling — and it’s overall significance — which you can find here.)
(2) Orrick Isn’t Alone. More Judges Are Going His Way.
Second, Judge Orrick isn’t alone anymore in the way he’s thinking about “fair use” in the context of GenAI. At least two other federal judges presiding over GenAI infringement cases also recently kept the issue of “fair use” alive as a matter of law — and concluded that they too can decide the issue pre-trial on summary judgment — Judge Jon S. Tigar (U.S. District Court, Northern District of California) and Judge Stephanos Bibas (U.S. District Court, Delaware) (NOTE: I recently wrote an entire analysis of these rulings — and their overall significance — which you can find here.)
And also recently, another federal Judge (Vince Chhabria, Northern District of California), seemingly sympathetic to rights-holders, essentially pulled the rip-chord on their lead attorney in several closely watched consolidated cases for not effectively representing them. Enter famed litigator David Boies who certainly will be hammering home these “fair use” arguments on behalf of these plaintiffs, who include comedian Sarah Silverman.
(3) When These Courts Decide This “Fair Use” Issue, They Have New Supreme Court Precedent to Back Them Up.
Third, The federal courts and judges noted above have already concluded that the question of infringement in the generative AI context has two parts – GenAI training (inputs) and prompt outputs. And recent Supreme Court precedent points to a rejection of Big Tech "fair use" defenses on the question of AI training (the input side of the equation). Wholesale copying, together with market substitution, is enough to find infringement.
Andy Warhol v. Goldsmith: New Supreme Court Precedent
In its most recent infringement case (Andy Warhol v. Goldsmith), the Supreme Court didn’t even reach the issue of infringement on the output side. The plaintiff (a professional photographer) had already conceded that issue and agreed that the allegedly infringing works — Warhol’s series of silkscreen prints and paintings of musician Prince based on the relevant photograph — were "transformative.” Instead, the Court focused only on how the creative work (the “input” photograph) was used.
In rejecting the infringer’s “fair use” argument, the 7-2 majority ruled that its works’ “transformative” nature was not enough. It noted that copyright’s protection is even stronger “where the copyrighted material serves an artistic rather than utilitarian function" — and further that market “substitution” is “copyright’s bete noire.” The Court agreed with the plaintiff photographer that Warhol’s foundation (the defendant) usurped a commercial opportunity by not seeking a license from her. The Court wrote that “the use of an original work to achieve a purpose that is the same as, or highly similar to, that of the original work is more likely to substitute for, or supplant, the work” which, in turn, “undermines the goal of copyright.”
Google Books: Big Tech’s “Go To” Fair Use Case Falls Flat.
Big Tech always points to the "Google Books" case decided by the Second Circuit Court of Appeals to support its fair use defense in the GenAI context. In that case, Google copied the world’s library of books word-for-word to create a searchable online index — and then would display small “snippets” of those works (not the entirety of those books) — in its search results. In that specific context, the court ruled in Google’s favor to reject infringement.
But “Google Books” was a Second Circuit opinion - not the Supreme Court's law of the land. But even if it were, the Google Books court's core justification was market amplification (not substitution or replacement). Google's wholesale copying in that context shined a brighter light on those copied books — more attention and monetization to Creators, not less.
It's precisely the opposite when AI scrapes creative works. Generative AI creates commercial substitutes for wholesale sectors of news, media & entertainment. Big Tech’s aim is frequently to be a one-stop-shop replacement for those industries’ works, and the Supreme explicitly rejected reliance on Google Books for similar reasons. The 7-2 majority emphasized that context is critical — using the Second Circuit’s earlier decision in the widely cited American Geophysical Union v. Texaco case to support its own “fair use” rejection. Texaco’s copying in that case had a direct and negative impact on the market for the scientific articles at issue there, because its copying allowed Texaco to avoid purchasing additional subscriptions or licenses. In other words, direct commercial harm.
(You can read my earlier full discussion of the Supreme Court, its Andy Warhol copyright ruling, and “Google Books” here.)
(4) And Now California Has New GenAI “Transparency” Law.
Fourth, California Governor Gavin Newsom just signed 17 new GenAI bills into law, including AB 2013, which requires so-called “transparency” — forcing the major GenAI developers (OpenAI, Meta, Google, Microsoft, Amazon, Apple) to reveal the content sources they use to train their AI models (you can find my full discussion and analysis of it all here). That spotlight puts even more pressure on these companies to do it right. They’ll no longer be able to simply hide behind those three words again (“publicly available data”) in their transparent efforts to obfuscate.
(5) It’s A Big World. “Fair Use” Here, May Not Be “Fair” Over There.
Fourth, let’s not forget it’s a big world out there. So even if GenAI developers ultimately prevail on “fair use” arguments in the U.S. courts, that certainly doesn’t end the issue in the 200+ other nations out there (most of which have their own copyright schemes). GenAI tools are borderless, of course. That means that litigation can happen anywhere.
What solves this dilemma? Global licensing deals with copyright owners!
(6) “Good Faith” Matters. And Frequently There Isn’t Any.
Finally, U.S. courts apply a four-part balancing test when analyzing “fair use” which considers, among other things, “the purpose and character of the use” — including whether the challenged use is commercial or nonprofit). Although an alleged infringer’s “good faith” is not a formal element of this four-part test, a lack of good faith certainly can influence a court’s decision.
And so, let’s look at some of the leading alleged infringers — and the type of “faith” they’ve demonstrated — in this context. I’ll give you two examples to show how many GenAI developers and Silicon Valley feel about content (and value the creators behind their creative works).
Exhibit A: Mark Zuckerberg and Meta. When Meta had fallen behind OpenAI on the GenAI front, CEO Zuckerberg gave the edict to move fast to catch up. According to The New York Times, company stakeholders internally cautioned that consent would be needed to scrape copyrighted work. But the decision was made to move forward anyhow without licensing because it simply would be too difficult and time-consuming to do otherwise. In other words, break first, ask for forgiveness later. Zuckerberg just recently kicked more sand in the face of rights-holders as he arrogantly dismissed the value of their contributions to his AI tech. That’s rich, coming from a guy whose GenAI tech would have zero value without the content its AI models needed.
Exhibit B: Venture Capitalist Antonio Rodriguez and Suno. Suno is a leading generative music company that also finds itself in the legal crosshairs. It’s certainly powerful, enabling anyone to create professionally sounding music tracks in seconds simply by text prompting. The problem is that, just like the Big Tech giants it emulates, Suno trains its AI on endless streams of copyrighted music without consent. It finally conceded that inconvenient truth in court filings (extremely brashly, by the way), only when it was forced to. Before being sued, Suno simply would use — you guessed it — those three words (“publicly available data”) when asked about what content it used for training its AI model).
Here’s the “not so good faith” part. One of the company’s early investors, Antonio Rodriguez of Matrix Partners, notoriously said the quiet part out loud when asked months before litigation was filed. In his words, “Honestly if we had deals with labels when this company got started, I probably wouldn’t have invested in it. I think that they needed to make this product without the constraints.” (NOTE: you can read my earlier full analysis about it all here).
Zuckerberg and Rodriguez pretty much sum it up for many (if not most) generative AI developers and their investors – who find dealing with copyrights and licensing to be needless, pesky “constraints.”
To Be Clear …
None of this means that it’s a slam dunk that the courts will reject “fair use” in the context of GenAI training on copyrighted works without consent and compensation. But, taken together, this growing list of data points reflect a trend that I believe favors rights-holders.
In other words, Silicon Valley’s “fair use” dominoes are beginning to fall. And the best way to stack them back up is cooperatively with other hands — rights-holder hands — and together move beyond “fair use” litigation to actual paid fair use and fair licensing of their content.
Listen to a Smart Conversation About My Article - and the Issues It Raises - On This Week’s Podcast
This episode is surprisingly comprehensive, insightful and entertaining — check it out (I generated it by simply dropping the article’s text into Google NotebookLM. No prompting needed).
What do you think? Drop me a line at peter@creativemedia.biz.
Last Week’s Generative AI Event Highlight
This past Friday, I was asked by Google to lead an impressive industry panel in advance of LA Tech Week. It was a wonderful event where I shared the stage with Paul Trillo (whose AI generated film is highlighted above), Jenn Haran of YouTube, Clarence Wooten of Google’s “moonshot” division X, and AI LA’s Todd Terrazes to discuss GenAI’s impact on the creative community (and media and entertainment in general). Here’s a scene.
II. GenAI Video of the Week
Here is award-winning GenAI filmmaker Paul Trillo’s latest GenAI-created film “Etherea.” Just press play to watch it. You’ll see what I mean ….
III. AI Litigation Case Tracker - Updates on Key AI Litigation (brought to you by McKool Smith)
Partner Avery Williams and the team at McKool Smith (recently named “Plaintiff IP Firm of the Year”) lay out the facts - and latest critical developments - via this link to the “AI Litigation Tracker”. You’ll get everything you need (including my own separate detailed analyses) for each of the cases listed below. So much you need to know (no matter what role you play). So little time. We do the work so you can relax!
Here’s the central theme in all these cases: GenAI developers need access to the world of content for training their AI models (it’s all just “data” to them). But copyright owners understandably believe (as do I) that those tech companies must seek consent from — and pay compensation to — the relevant content/copyright owners whose content they scrape. And when they don’t, they increasingly find themselves in court.
(1) The New York Times v. Microsoft & OpenAI
(2) Sarah Silverman v. OpenAI (class action)
(3) Sarah Silverman, et al. v. Meta (class action)
(4) UMG Recordings v. Suno
(5) UMG Recordings v. Uncharted Labs (d/b/a Udio)
(6) Getty Images v. Stability AI and Midjourney
(7) Universal Music Group, et al. v. Anthropic
(8) Sarah Anderson v. Stability AI
(9) Authors Guild et al. v. OpenAI
NOTE: Go to the “AI Litigation Tracker” tab at the top of “the brAIn” website for the full discussions and analyses of these and other key generative AI/media litigations. And reach out to me, Peter Csathy (peter@creativemedia.biz), if you would like to be connected to McKool Smith) to explore these and other legal and litigation issues. I’ll make the introduction
And check out my firm Creative Media and our AI-focused services
Send feedback to my newsletter - including guest essay submissions and other ideas - to me, Peter Csathy, at peter@creativemedia.biz.