The Times v. Microsoft/OpenAI: A Business Model Based on Mass Copyright Infringement (8)

IV. FACTUAL ALLEGATIONS

B. Defendants’ GenAI Products

1. A Business Model Based on Mass Copyright Infringement

55. OpenAI was formed in December 2015 as a “non-profit artificial intelligence research company.” OpenAI started with $1 billion in seed money from its founders, a group of some of the wealthiest technology entrepreneurs and investors and companies like Amazon Web Services and InfoSys. This group included Elon Musk, the CEO of Tesla and X Corp. (formerly known as Twitter); Reid Hoffman, the co-founder of LinkedIn; Sam Altman, the former president of Y Combinator; and Greg Brockman, the former Chief Technology Officer of Stripe.

56. Despite accepting very large investments from enormously wealthy companies and individuals at its founding, OpenAI originally maintained that its research and work would be entirely unmotivated by profit. In a December 11, 2015, press release, Brockman and co-founder lya Sutskever (now OpenAI’s President and Chief Scientist, respectively) wrote: “Our goal is to advance digital intelligence in the way that is most likely to benefit humanity as a whole, unconstrained by a need to generate financial return. Since our research is free from financial obligations, we can better focus on a positive human impact.” In accordance with that mission, OpenAI promised that its work and intellectual property would be open and available to the public, that its “[r]esearchers will be strongly encouraged to publish their work, whether as papers, blog posts, or code” and that its “patents (if any) will be shared with the world.”

57. Despite its early promises of altruism, OpenAI quickly became a multi-billiondollar for-profit business built in large part on the unlicensed exploitation of copyrighted works belonging to The Times and others. Just three years after its founding, OpenAI shed its exclusively nonprofit status. It created OpenAI LP in March 2019, a for-profit company dedicated to conducting the lion’s share of OpenAI’s operations—including product development—and to raising capital from investors seeking a return. OpenAI’s corporate structure grew into an intricate web of for-profit holding, operating, and shell companies that manage OpenAI’s day-to-day operations and grant OpenAI’s investors (most prominently, Microsoft) authority and influence over OpenAI’s operations, all while raising billions in capital from investors. The result: OpenAI today is a commercial enterprise valued as high as $90 billion, with revenues projected to be over $1 billion in 2024.

58. With the transition to for-profit status came another change: OpenAI also ended its

commitment to openness. OpenAI released the first two iterations of its flagship GenAI model,

GPT-1 and GPT-2, on an open-source basis in 2018 and 2019, respectively. But OpenAI changed

course in 2020, starting with the release of GPT-3 shortly after OpenAI LP and other for-profit

OpenAI entities were formed and took control of product design and development.

59. GPT-3.5 and GPT-4 are both orders of magnitude more powerful than the two previous generations, yet Defendants have kept their design and training entirely a secret. For previous generations, OpenAI had voluminous reports detailing the contents of the training set, design, and hardware of the LLMs. Not so for GPT-3.5 or GPT-4. For GPT-4, for example, the “technical report” that OpenAI released said: “this report contains no further details about the architecture (including model size), hardware, training compute, dataset construction, training method, or similar.”[3]

60. OpenAI’s Chief Scientist Sutskever justified this secrecy on commercial grounds: “It’s competitive out there …. And there are many companies who want to do the same thing, so from a competitive side, you can see this as maturation of the field.”[4] But its effect was to conceal the identity of the data OpenAI copied to train its latest models from rightsholders like The Times.

61. OpenAI became a household name upon the release of ChatGPT in November 2022. ChatGPT is a text-generating chatbot that, given user-generated prompts, can mimic humanlike natural language responses. ChatGPT was an instant viral sensation, reaching one million users within a month of its release and gaining over 100 million users within three months.

62. OpenAI, through OpenAI OpCo LLC and at the direction of OpenAI Inc., OpenAI LP, and other OpenAI entities, offers a suite of services powered by its LLMs, targeted to both ordinary consumers and businesses. A version of ChatGPT powered by GPT-3.5 is available to users for free. OpenAI also offers a premium service, powered by OpenAI’s “most capable model” GPT-4, to consumers for $20 per month. OpenAI’s business-focused offerings include ChatGPT Enterprise and ChatGPT API tools designed to enable developers to incorporate ChatGPT into bespoke applications. OpenAI also licenses its technology to corporate clients for licensing fees.

63. These commercial offerings have been immensely valuable for OpenAI. Over 80% of Fortune 500 companies are using ChatGPT.[5] According to recent reports, OpenAI is generating revenues of $80 million per month, and is on track to surpass over $1 billion within the next 12 months.[6]

64. This commercial success is built in large part on OpenAI’s large-scale copyright infringement. One of the central features driving the use and sales of ChatGPT and its associated products is the LLM’s ability to produce natural language text in a variety of styles. To achieve this result, OpenAI made numerous reproductions of copyrighted works owned by The Times in the course of “training” the LLM.

65. Upon information and belief, all of the OpenAI Defendants have been either directly involved in or have directed, controlled, and profited from OpenAI’s widespread infringement and commercial exploitation of Times Works. OpenAI Inc., alongside Microsoft, controlled and directed the widespread reproduction, distribution, and commercial use of The Times’s material perpetrated by OpenAI LP and OpenAI Global LLC, through a series of holding and shell companies that include OpenAI Holdings LLC, OpenAI GP LLC, and OAI Corporation LLC. OpenAI LP and OpenAI Global LLC were directly involved in the design, development, and commercialization of OpenAI’s GPT-based products, and directly engaged in the widespread reproduction, distribution, and commercial use of Times Works. OpenAI LP and OpenAI Global LLC also controlled and directed OpenAI, LLC and OpenAI OpCo LLC, which were involved in distributing, selling, and licensing OpenAI’s GPT-based products, and thus monetized the reproduction, distribution, and commercial use of Times Works.

66. Since at least 2019, Microsoft has been, and continues to be, intimately involved in the training, development, and commercialization of OpenAI’s GPT products. In an interview with the Wall Street Journal at the 2023 World Economic Forum, Microsoft CEO Satya Nadella said that the “ChatGPT and GPT family of models … is something that we’ve been partnered with OpenAI deeply now for multiple years.” Through this partnership, Microsoft has been involved in the creation and commercialization of GPT LLMs and products based on them in at least two ways.

67. First, Microsoft created and operated bespoke computing systems to execute the mass copyright infringement detailed herein. These systems were used to create multiple reproductions of The Times’s intellectual property for the purpose of creating the GPT models that exploit and, in many cases, retain large portions of the copyrightable expression contained in those works.

68. Microsoft is the sole cloud computing provider for OpenAI. Microsoft and OpenAI collaborated to design the supercomputing systems powered by Microsoft’s cloud computer platform Azure, which were used to train all OpenAI’s GPT models after GPT-1. In a July 2023 keynote speech at the Microsoft Inspire conference, Mr. Nadella said: “We built the infrastructure to train their models. They’re innovating on the algorithms and the training of these frontier models.”

69. That infrastructure was not just general purpose computer systems for OpenAI to use as it saw fit. Microsoft specifically designed it for the purpose of using essentially the whole internet—curated to disproportionately feature Times Works—to train the most capable LLM in history. In a February 2023 interview, Mr. Nadella said:

But beneath what OpenAI is putting out as large models, remember,

the heavy lifting was done by the [Microsoft] Azure team to build

the computer infrastructure. Because these workloads are so

different than anything that’s come before. So we needed to

completely rethink even the datacenter up to the infrastructure that

first gave us even a shot to build the models. And now we’re

translating the models into products.[7]

70. Microsoft built this supercomputer “in collaboration with and exclusively for OpenAI,” and “designed [it] specifically to train that company’s AI models.”[8] Even by supercomputing standards, it was unusually complex. According to Microsoft, it operated as “a single system with more than 285,000 CPU cores, 10,000 GPUs and 400 gigabits per second of network connectivity for each GPU server.” This system ranked in the top five most powerful publicly known supercomputing systems in the world.

71. To ensure that the supercomputing system suited OpenAI’s needs, Microsoft needed to test the system, both independently and in collaboration with OpenAI software engineers. According to Mr. Nadella, with respect to OpenAI: “They do the foundation models, and we [Microsoft] do a lot of work around them, including the tooling around responsible AI and AI safety.” Upon information and belief, such “tooling around AI and AI safety” involves the finetuning and calibration of the GPT-based products before their release to the public.[9]

72. In collaboration with OpenAI, Microsoft has also commercialized OpenAI’s GPTbased technology, and combined it with its own Bing search index. In February 2023, Microsoft unveiled Bing Chat, a generative AI chatbot feature on its search engine powered by GPT-4. In May 2023, Microsoft and OpenAI unveiled “Browse with Bing,” a plugin to ChatGPT that enabled it to access the latest content on the internet through the Microsoft Bing search engine. Bing Chat and Browse with Bing combine GPT-4’s ability to mimic human expression—including The Times’s expression—with the ability to generate natural language summaries of search result contents, including hits on Times Works, that obviate the need to visit The Times’s own websites. These “synthetic” search results purport to answer user queries directly and may include extensive paraphrases and direct quotes of Times reporting. Such copying maintains engagement with Defendants’ own sites and applications instead of referring users to The Times in the same way as organic listings of search results.

73. In a recent interview, Mr. Nadella acknowledged Microsoft’s intimate involvement in OpenAI’s operations and, therefore, its copyright infringement:

[W]e were very confident in our own ability. We have all the IPrights and all the capability. If OpenAI disappeared tomorrow, Idon’t want any customer of ours to be worried about it quite honestly, because we have all of the rights to continue theinnovation. Not just to serve the product, but we can go and just dowhat we were doing in partnership ourselves. We have the people,we have the compute, we have the data, we have everything.

74. Through their collaboration in both the creation and the commercialization of the GPT models, Defendants have profited from the massive copyright infringement, commercial exploitation, and misappropriation of The Times’s intellectual property. As Mr. Nadella recently put it, “[OpenAI] bet on us, we bet on them.” He continued, describing the effect of Microsoft’s $13 billion investment:

And that gives us significant rights as I said. And also this thing, it’snot hands off, right? We are in there. We are below them, abovethem, around them. We do the kernel optimizations, we build tools,we build the infrastructure. So that’s why I think a lot of theindustrial analysts are saying, ‘Oh wow, it’s really a joint projectbetween Microsoft and OpenAI.’The reality is we are, as I said, very self-sufficient in all of this.

[3] OPENAI, GPT-4 TECHNICAL REPORT (2023), https://cdn.openai.com/papers/gpt-4.pdf.

[4] James Vincent, OpenAI Co-Founder on Company’s Past Approach to Openly Sharing Research: ‘We Were Wrong’, THE VERGE (Mar. 15, 2023), https://www.theverge.com/2023/3/15/23640180/openai-gpt-4-launch-closedresearch-ilya-sutskever-interview.

[5] OpenAI, Introducing ChatGPT Enterprise, OPENAI (Aug. 28, 2023), https://openai.com/blog/introducing-chatgpt-enterprise.

[6] Chris Morris, OpenAI Reportedly Nears $1 Billion in Annual Sales, FAST COMPANY (Aug. 30, 2023), https://www.fastcompany.com/90946849/openai-chatgpt-reportedly-nears-1-billion-annual-sales.

[7] First on CNBC: CNBC Transcript: Microsoft CEO Satya Nadella Speaks with CNBC’s Jon Fortt on

“Power Lunch” Today, CNBC (Feb. 7, 2023), https://www.cnbc.com/2023/02/07/first-on-cnbc-cnbc-transcriptmicrosoft-ceo-satya-nadella-speaks-with-cnbcs-jon-fortt-on-power-lunch-today.html.

[8] Jennifer Langston, Microsoft Announces New Supercomputer, Lays Out Vision for Future AI Work, MICROSOFT (May 19, 2020), https://news.microsoft.com/source/features/ai/openai-azure-supercomputer/. 9 SÉBASTIEN BUBECK ET AL., SPARKS OF ARTIFICIAL GENERAL INTELLIGENCE: EARLY EXPERIMENTS WITH GPT-4 (2023), https://arxiv.org/pdf/2303.12712.pdf