Context: In late December. the New York Times sued Microsoft and OpenAI over ChatGPT’s use of some of its copyrighted material (case 1:23-cv-11195, Southern District of New York) (first ai fray article on this dispute). Fair use will be the key defense, and proportionality considerations are key to three of the fair use factors as well as any hypothetical remedies (December 29 article).
What’s new: Last week, Orrick Herrington & Sutcliffe partner Annette Hurst (profile on firm’s website) entered her appearance (PDF) as Microsoft’s counsel.
Direct impact: While Microsoft is the first-named defendant, the case is actually about OpenAI’s technology and its Large Language Model (LLM) training. OpenAI already explained in a January 8 blog post the background of, and its perspective on, this case. But the New York Times presumably hopes to capitalize on Microsoft’s size in the damages context. What makes Microsoft’s choice of Mrs. Hurst (who has previously worked for the company, such as in a Github case) particularly interesting is that she was on the receiving end of the Supreme Court’s Oracle v. Google fair use ruling.
Wider ramifications: This litigation, unless it is settled on the basis of a license agreement, has the potential to become the next major U.S. fair use case. The Supreme Court’s Oracle v. Google opinion (PDF) will presumably be cited extensively as it was the last major case in which the top U.S. court stressed the need to “keep the copyright monopoly […] within its lawful bounds” and that “creative progress […] is the basic constitutional objective of copyright itself.” It was also resolved definitively in Oracle v. Google that “[t]he ‘right of trial by jury’ does not include the right to have a jury resolve a fair use defense” (the NYT is seeking a jury trial, but fair use will be decided by the judge, so a jury will only get involved if the NYT can overcome the fair use defense).
In some ways, litigation is like a team sport, like a ball game. But not in every way Applying the simpler logic of the sports world, the question would inevitably be why Microsoft would hire a lawyer who lost the last high-profile fair use case. It might look like a team aspiring to win the Super Bowl hiring the coach who lost the last one. And even in sports, there have been cases where losses were painful in the immediate term, but instructive (like a tuition fee) and the lessons learned enabled a major success.
What decided Oracle v. Google was not the quality of the legal teams. In fact, some of those who wanted Google to win commented very negatively on the performance of Google’s Supreme Court counsel after the hearing (which didn’t appear to go well for Google). That same lawyer, by the way, lost Epic v. Apple in the Ninth Circuit, but turned around FTC v. Qualcomm. The dissent filed by Justices Thomas and Alito essentially says the majority just wanted a particular fair use outcome (and the two judges in the minority sharply disagreed with how the majority got there).
Why was Oracle v. Google unwinnable in the end? That question is interesting way beyond the fact that Microsoft brought Oracle’s then-losing law firm on board. The case shows how difficult it is to overcome a fair use defense when there’s a strong interest in a certain new way to use copyrighted material being held lawful.
That litigation lasted more than a decade (2010-2021), requiring two jury trials, after each of which the United States Court of Appeals for the Federal Circuit overruled the district court. There will be more opportunities throughout New York Times v. Microsoft & OpenAI where Oracle v. Google is going to come up and various aspects can be discussed, but this article serves the purpose of providing a general overview of the arguments, procedural steps and circumstances that led to the decision. First, the key facts and the procedural steps in a very compressed form:
- Google designed Android’s application programming interface (API) (i.e., the part of the program code that defines how Android applications make use of the functionality of the underlying system, such as by displaying a message or initiating a phone call) in a way that app developers already familiar with Oracle’s Java programming language (which belonged to Sun Microsystems at the time, a company Oracle later acquired for $7 billion) would easily find their way around. Android was never 100% Java-compatible, but the common elements simplified things for app makers and helped Android bring developers on board early on. It can’t be ruled out, but it’s far from a given, that Android would have succeeded even with an entirely novel, independently-created API.
- In order to benefit from such familiarity, Google copied roughly 7,000 lines of source code. Compared to the 300-400 words of President Ford’s book manuscript that the Supreme Court thought were not used fairly in the Harper & Row case, that’s a huge amount. Compared to Google Books, it’s obviously nothing. There are also some cases were huge quantities of text were legally copied, but those cases involved non-creative material (such as phone directories).
- Oracle’s argument was that they took the most valuable part and one that involved a particularly high degree of creative expression (which copyright is about). As stated above, Google benefited hugely, and effectively drove Java out of the mobile business (a negative impact on the original work that normally weighs very much against a finding of fair use).
- Google’s argument was that they took just (part of) the interfaces but did the entire implementation (tens of millions of lines of code) independently. It’s not even like fleshing out a skeleton; it was more like they took some of the contours of a city from a postcard and then designed, built, furnished and populated an entire city themselves.
- Oracle asserted primarily patents and, what was initially probably just an afterthought or at least a lower priority, also copyrights. There is no fair use defense to patent infringement, so if Oracle had prevailed on one or more of its patents, and depending on whether it would have been technically possible and economically viable for Google to work around them, Oracle could have had all the leverage it needed to force Google into a costly settlement.
- Google won the first round because the district court interpreted certain Ninth Circuit copyright rulings related to video game consoles (Sony and Sega) as making API code non-copyrightable. The Federal Circuit reversed. But the first jury was “hung” (they couldn’t agree) on fair use. The Federal Circuit didn’t want to resolve fair use immediately, particularly because it had rejected the non-copyrightability defense on the grounds that the Sony and Sega holdings had to be relegated to the fair use part (a subsequent question).
- Mrs. Hurst was then Oracle’s lead counsel in the retrial. The retrial jury, instructed by a judge who didn’t make it easy for Oracle to win, said it was fair use. The Federal Circuit got involved again and overruled once more, holding that what Google had done was clearly unfair. The Supreme Court then (though it wasn’t interested in the copyrightability part) accepted the case and not only held that there was nothing unreasonable about the jury verdict but even went beyond what Google itself had said and held that this was fair use, no question, end of story.
There was nothing about Oracle’s arguments in that case that suggested an error, oversight or lack of persuasiveness by its legal team. No commentator criticized the quality of their work, while (as mentioned further above) some of Google’s supporters were actually displeased with the performance of Google’s Supreme Court counsel. Also, the latter made various highly questionable claims, such as saying that Oracle hadn’t moved for summary judgment, when the record proves that the district judge simply didn’t allow it when it mattered. By contrast, Oracle’s lawyers were truthful, and while the ultimate outcome was bad, they had won twice in the Federal Circuit (i.e., they revived a case that was dead after a district court judgment), which was nothing short of remarkable.
What happened between the Supreme Court hearing and the decision was an about-face. At the hearing, a majority of the judges was really unconvinced of Google’s fair use arguments. That was the more traditional stance. Up to that point, it was deemed fair use to copy large quantities of program code only for internal purposes, such as for the purpose of then being able to develop something new. One could also copy small amounts such as a 20-byte sequence and incorporate them into widely-distributed products. But the combination of 7,000 lines of code and many billions of copies (just think of all those Android devices out there over all those years) was unprecedented.
If one reads the per curiam (the majority opinion), it is reasonable to deduce or speculate that the Chief Judge and six of the Associate Justices were driven by two considerations:
- Oracle was seeking to overleverage its API code, trying to get a multi-billion payout by potentially destabilizing an entire mobile ecosystem if Google wasn’t going to pay.
- Software copyright as a whole wouldn’t be weakened by making it clear that this is a highly case-specific determination and that API code should be treated differently in the fair use context than other (actually implementing) code.
So they felt they could protect regular program code against copying while also protecting innovation through independent implementations. They could achieve something positive without a significant risk of adverse effects. That is, by the way, also the case in that OpenAI case: innovation can be supported without doing something that means anyone could just publish on the open internet the NYT’s paywalled articles.
The dissent makes very good points as to why that departure from traditional fair use law was a bad precedent and weakened copyright protection. It’s also perfectly plausible that the majority decided not to address copyrightability (they said they didn’t have to because it was fair use anyway) because that would have required them to face the fact that API code is particularly creative (even a Google witness said so under oath). At the hearing, that traditional perspective appeared to be almost everyone’s line (other than maybe Justice Sotomayor’s; Justice Breyer wrote the per curiam and both may have been very persuasive in post-hearing deliberations and discussions).
Given that both the majority and the minority made valid points, it was a difficult call, even more difficult than the 6-2 vote (at the time of the hearing, the Supreme Court was one judge short) suggests. There was one factor here that no one could ever blame on Mrs. Hurst and her colleagues: Google, with its vast resources, ran a massive amicus brief campaign. Not just amicus briefs but also academic writings and articles. It was a full-blown campaign, and they didn’t even need it to win the hearts and minds of most software developers.
There was one key misconception that benefited Google:
- It is obviously key for developers to ensure their access to APIs in terms of writing applications that use an API. For example, if Oracle had at some point said that all Java programs are now illegal without a special license that costs some extra money, that would have impacted the developer ecosystem.
- But hardly a developer (and especially no small or medium-sized company, much less an individual programmer) does what Oracle was really going against in this case: it was about (sort of) cloning an existing system, thereby just benefiting from what Oracle (or, as mentioned further above, the company it acquired) had done to make Java so popular. That takes enormous resources.
- Even that part was not an issue in the case at hand as Oracle offered two other ways in which Google could have used even the entire Java API code: it could have used it under an open-source license requiring it to contribute its own derivative works back to the ecosystem, and it could have taken a license on fair, reasonable and non-discriminatory (FRAND) terms without those “copyleft” (contribute-back) obligations, though full code compatibility would have been a requirement in the latter case.
In the public debate, those two types of use (making apps vs. making a competing platform) were conflated all the time. A lot of people thought that a victory for Oracle could have exposed them personally to liability when there was never any such risk. The Supreme Court may not have fully understood that most of the doomsday predictions submitted to it in the form of amicus briefs blew things far out of proportion. They just saw there was widespread concern, and they found a way to assuage those concerns without abolishing software copyright or weakening it substantially.
By now, Google is more controversial than it was in the early years of the dispute (i.e., the early 2010s), which was, however, when many people formed their opinion. Oracle was never the most well-liked tech company. Whenever Google won something during the process, one could also see comments on social media that targeted Oracle founder Larry Ellison personally. Later on, Google had become way larger and influential than Oracle. It had a far more powerful contact network and leveraged it to influence the Supreme Court.
It wouldn’t have been feasible for Oracle to counterbalance Google’s campaign in full. They could have done more than they did to explain to the developer community what it’s really about, but it was a costly 11-year dispute and less than two years after filing the complaint they suffered a couple of near-simultaneous setbacks with the jury being unable to decide fair use (apparently all jurors except the foreman were on Google’s side) and the judge even holding the API code uncopyrightable. They carried on, but probably weren’t going to spend too much money and also weren’t going to allocate too much management bandwidth.
One thing they arguably could or should have done differently is that they shouldn’t have limited themselves to just stressing (in other words) that “the law favors Oracle because fair use is a rather narrow exception and copyright infringement is the rule.” What Oracle’s lawyers failed to do was to explain to the courts, especially to the Supreme Court, that any concerns about overleveraging could be addressed differently than just expanding the fair use doctrine.
They solved the copyrightability problem because the appeals court felt that the related arguments were fair use, not non-copyrightability arguments. But Oracle didn’t even make an attempt to convince a court that even fair use is not the end of the road and that those concerns over overleveraging of API code could be relegated once more. They could have said that if all else fails and someone really abuses API copyrights, there’s antitrust law.
For a duty to deal (meaning that someone with market power is forced to do business with one or more third parties on certainterms), U.S. antitrust law has a very high hurdle. But if someone abused API copyrights with a bait-and-switch strategy (first letting developers use a free API and then turning around), the Aspen Skiing criteria (paticularly the one of abandoning a prior profitable course of action) could be met. Oracle, however, didn’t say: let us win the copyright part because in our case, there are different licensing options for Google anyway, and if they didn’t exist or if anything ever changed about that, Google could still try to bring competition claims.
Whether that would have changed the outcome is another question. It is, however, true that Oracle only had traditional fair use law on its side and didn’t propose anything other than rigid copyright enforcement to address any policy concerns. Google prevailed on policy (how to spur innovation and creativity), politics (far, far more support from the ecosystem) and psychology (while fair use law formally doesn’t have a “most of what we do is lawful” defense, judges and jurors are just human and if they see that someone took very little in order to build something huge that is new and valuable, it doesn’t feel right then to say “it’s illegal and there must be draconian consequences”).
The New York Times now has to overcome a fair use defense, and if OpenAI is right that “regurgitation” of entire paragraphs from NYT articles is a “rare bug” they’re trying to fix anyway, then the NYT case will be seen as an even more extreme case of overleveraging that Oracle’s enforcement of API copyrights.
In another high-profile context that involves Microsoft, lawyers actually benefited from a decision in a case their firm had lost a few months before. In 2012, Microsoft won an antisuit injunction (more specifically, an anti-enforcement injunction that barred Motorola Mobility, which was at the time being acquired by Google, from enforcing two German patent injunctions) in the Western District of Washington (Judge James L. Robart) that was upheld by the United States Court of Appeals for the Ninth Circuit. Motorola was then represented by Quinn Emanuel. A few years later, QE won a U.S. antisuit injunction against Huawei in the Northern District of California (Judge William H. Orrick), pointing to the very case they had lost against Microsoft. And while it’s unlikely that any of Google’s lawyers will work for the NYT now, it’s a major coincidence that Huawei was represented by Sidley Austin, a firm that frequently works for Microsoft and had won that antisuit injunction against Motorola. On Huawei’s behalf they became a victim of their success on Microsoft’s behalf.
Against the NYT’s complaint, OpenAI and Microsoft have until March 4 (within 60 days of formal service) to respond. They can file an answer to the complaint or bring a motion to dismiss. Given that the NYT’s legal theories are ambitious, a motion to dismiss is anything but unlikely to be brought, as it might narrow the case (even if only gradually).