Case 4:23-cv-03416-KAW Document 1 Filed 07/07/23 Page 1 of 17
`
`
`
`Joseph R. Saveri (State Bar No. 130064)
`Cadio Zirpoli (State Bar No. 179108)
`Christopher K.L. Young (State Bar No. 318371)
`Kathleen J. McMahon (State Bar No. 340007)
`JOSEPH SAVERI LAW FIRM, LLP
`601 California Street, Suite 1000
`San Francisco, California 94108
`Telephone:
`(415) 500-6800
`Facsimile:
`(415) 395-9940
`Email:
`jsaveri@saverilawfirm.com
`czirpoli@saverilawfirm.com
`cyoung@saverilawfirm.com
`kmcmahon@saverilawfirm.com
`
`
`Matthew Butterick (State Bar No. 250953)
`1920 Hillhurst Avenue, #406
`Los Angeles, CA 90027
`Telephone:
`(323) 968-2632
`Facsimile:
`(415) 395-9940
`mb@buttericklaw.com
`Email:
`
`Counsel for Individual and Representative Plaintiffs
`and the Proposed Class
`
`UNITED STATES DISTRICT COURT
`NORTHERN DISTRICT OF CALIFORNIA
`SAN FRANCISCO DIVISION
`
`
`Sarah Silverman, an individual;
`Christopher Golden, an individual;
`Richard Kadrey, an individual;
`
`Individual and Representative Plaintiffs,
`
`v.
`
`OpenAI, Inc., a Delaware nonprofit corporation; OpenAI, L.P., a
`Delaware limited partnership; OpenAI OpCo, L.L.C., a Delaware
`limited liability corporation; OpenAI GP, L.L.C., a Delaware
`limited liability company; OpenAI Startup Fund GP I, L.L.C.,
`a Delaware limited liability company; OpenAI Startup Fund I,
`L.P., a Delaware limited partnership; and OpenAI Startup Fund
`Management, LLC, a Delaware limited liability company,
`
`Case No.
`
`Complaint
`
`Class Action
`
`Demand for
`Jury Trial
`
`
`
`
`
`Defendants.
`
`
`COMPLAINT
`
`
`
`1 2 3 4 5 6 7 8 9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25
`
`26
`
`27
`
`28
`
`

`

`Case 4:23-cv-03416-KAW Document 1 Filed 07/07/23 Page 2 of 17
`
`
`
`Plaintiffs Sarah Silverman, Christopher Golden, and Richard Kadrey (“Plaintiffs”), on behalf of
`
`themselves and all others similarly situated, bring this Class Action Complaint (the “Complaint”)
`
`against Defendants OpenAI, Inc., OpenAI, L.P., OpenAI OpCo, L.L.C., OpenAI GP, L.L.C., OpenAI
`
`Startup Fund I, L.P., OpenAI Startup Fund GP I, L.L.C. and OpenAI Startup Fund Management, LLC
`
`for direct copyright infringement, vicarious copyright infringement, violations of section 1202(b) of the
`
`Digital Millennium Copyright Act, unjust enrichment, violations of the California and common law
`
`unfair competition laws, and negligence. Plaintiffs seek injunctive relief an to recover damages as a
`
`result and consequence of Defendants’ unlawful conduct.
`I.
`
`OVERVIEW
`
`1.
`2.
`
`ChatGPT is a software product created, maintained, and sold by OpenAI.
`
`ChatGPT is powered by two AI software programs called GPT-3.5 and GPT-4, also
`
`known as large language models. Rather than being programmed in the traditional way, a large language
`
`model is “trained” by copying massive amounts of text and extracting expressive information from it.
`
`This body of text is called the training dataset. Once a large language model has copied and ingested the
`
`text in its training dataset, it is able to emit convincingly naturalistic text outputs in response to user
`
`prompts.
`3.
`
`A large language model’s output is therefore entirely and uniquely reliant on the
`
`material in its training dataset. Every time it assembles a text output, the model relies on the
`
`information it extracted from its training dataset.
`4.
`
`Plaintiffs and Class members are authors of books. Plaintiffs and Class members have
`
`registered copyrights in the books they published. Plaintiffs and Class members did not consent to the
`
`use of their copyrighted books as training material for ChatGPT. Nonetheless, their copyrighted
`
`materials were ingested and used to train ChatGPT.
`5.
`
`Indeed, when ChatGPT is prompted, ChatGPT generates summaries of Plaintiffs’
`
`copyrighted works—something only possible if ChatGPT was trained on Plaintiffs’ copyrighted works.
`6.
`
`Defendants, by and through the use of ChatGPT, benefit commercial and profit richly
`
`from the use of Plaintiffs’ and Class members’ copyrighted materials.
`
`1 2 3 4 5 6 7 8 9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25
`
`26
`
`27
`
`28
`
`
`
`
`
`1
`COMPLAINT
`
`
`
`

`

`Case 4:23-cv-03416-KAW Document 1 Filed 07/07/23 Page 3 of 17
`
`
`
`II.
`
`JURISDICTION AND VENUE
`
`7.
`
`This Court has subject matter jurisdiction under 28 U.S.C. § 1331 because this case
`
`arises under the Copyright Act (17 U.S.C. § 501) and the Digital Millennium Copyright Act (17 U.S.C.
`
`§ 1202).
`8.
`
`Jurisdiction and venue is proper in this judicial district under 28 U.S.C. § 1391(c)(2)
`
`because defendant OpenAI, Inc. is headquartered in this district, and thus a substantial part of the
`
`events giving rise to the claim occurred in this district; and because a substantial part of the events
`
`giving rise to Plaintiffs’ claims occurred in this District, and a substantial portion of the affected
`
`interstate trade and commerce was carried out in this District. Each Defendant has transacted business,
`
`maintained substantial contacts, and/or committed overt acts in furtherance of the illegal scheme and
`
`conspiracy throughout the United States, including in this District. Defendants’ conduct has had the
`
`intended and foreseeable effect of causing injury to persons residing in, located in, or doing business
`
`throughout the United States, including in this District.
`9.
`
`Under Civil Local Rule 3.2(c) and (e), assignment of this case to the San Francisco
`
`Division is proper because defendant OpenAI, Inc. is headquartered in San Francisco, a substantial
`
`amount part of the events giving rise to Plaintiffs’ claims and the interstate trade and commerce
`
`involved and affected by Defendants’ conduct giving rise to the claims herein occurred in this Division.
`III. PARTIES
`
`A.
`
`Plaintiffs
`10.
`
`Plaintiff Sarah Silverman is a writer and performer who lives in California. Plaintiff
`
`Silverman owns a registered copyright in one book, called The Bedwetter. This book contains copyright-
`
`management information customarily included in published books, including the name of the author
`
`and the year of publication.
`11.
`
`Plaintiff Christopher Golden is a writer who lives in Massachusetts. Mr. Golden owns
`
`registered copyrights in several books, including Ararat. This book contains the copyright-management
`
`information customarily included in published books, including the name of the author and the year of
`
`publication.
`
`
`
`
`
`2
`COMPLAINT
`
`
`
`1 2 3 4 5 6 7 8 9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25
`
`26
`
`27
`
`28
`
`

`

`Case 4:23-cv-03416-KAW Document 1 Filed 07/07/23 Page 4 of 17
`
`
`
`12.
`
`Plaintiff Richard Kadrey is a writer who lives in Pennsylvania. Plaintiff Kadrey owns
`
`registered copyrights in several books, including Sandman Slim. This book contains the copyright-
`
`management information customarily included in published books, including the name of the author
`
`and the year of publication.
`13.
`
`A nonexhaustive list of registered copyrights owned by Plaintiffs is included as
`
`Exhibit A.
`B.
`
`Defendants
`14.
`
`Defendant OpenAI, Inc. is a Delaware nonprofit corporation with its principal place of
`
`business located at 3180 18th St, San Francisco, CA 94110.
`15.
`
`Defendant OpenAI, L.P. is a Delaware limited partnership with its principal place of
`
`business located at 3180 18th St, San Francisco, CA 94110. OpenAI, L.P. is a wholly owned subsidiary
`
`of OpenAI Inc. that is operated for profit. OpenAI, Inc. controls OpenAI, L.P. directly and through the
`
`other OpenAI entities.
`16.
`
`Defendant OpenAI OpCo, L.L.C. is a Delaware limited liability company with its
`
`principal place of business located at 3180 18th Street, San Francisco, CA 94110. OpenAI OpCo,
`
`L.L.C. is a wholly owned subsidiary of OpenAI, Inc. that is operated for profit. OpenAI, Inc. controls
`
`OpenAI OpCo, L.L.C. directly and through the other OpenAI entities.
`17.
`
`Defendant OpenAI GP, L.L.C. (“OpenAI GP”) is a Delaware limited liability company
`
`with its principal place of business located at 3180 18th Street, San Francisco, CA 94110. OpenAI GP is
`
`the general partner of OpenAI, L.P. OpenAI GP manages and operates the day-to-day business and
`
`affairs of OpenAI, L.P. OpenAI GP was aware of the unlawful conduct alleged herein and exercised
`
`control over OpenAI, L.P. throughout the Class Period. OpenAI, Inc. directly controls OpenAI GP.
`18.
`
`Defendant OpenAI Startup Fund I, L.P. (“OpenAI Startup Fund I”) is a Delaware
`
`limited partnership with its principal place of business located at 3180 18th Street, San Francisco, CA
`
`94110. OpenAI Startup Fund I was instrumental in the foundation of OpenAI, L.P., including the
`
`creation of its business strategy and providing initial funding. OpenAI Startup Fund I was aware of the
`
`unlawful conduct alleged herein and exercised control over OpenAI, L.P. throughout the Class Period.
`
`
`
`
`
`3
`COMPLAINT
`
`
`
`1 2 3 4 5 6 7 8 9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25
`
`26
`
`27
`
`28
`
`

`

`Case 4:23-cv-03416-KAW Document 1 Filed 07/07/23 Page 5 of 17
`
`
`
`19.
`
`Defendant OpenAI Startup Fund GP I, L.L.C. (“OpenAI Startup Fund GP I”) is a
`
`Delaware limited liability company with its principal place of business located at 3180 18th Street, San
`
`Francisco, CA 94110. OpenAI Startup Fund GP I is the general partner of OpenAI Startup Fund I.
`
`OpenAI Startup Fund GP I is a party to the unlawful conduct alleged herein. OpenAI Startup Fund GP
`
`I manages and operates the day-to-day business and affairs of OpenAI Startup Fund I.
`20. Defendant OpenAI Startup Fund Management, LLC (“OpenAI Startup Fund
`
`Management”) is a Delaware limited liability company with its principal place of business located at
`
`3180 18th Street, San Francisco, CA 94110. OpenAI Startup Fund Management is a party to the
`
`unlawful conduct alleged herein. OpenAI Startup Fund Management was aware of the unlawful
`
`conduct alleged herein and exercised control over OpenAI, L.P. throughout the Class Period.
`IV. AGENTS AND CO-CONSPIRATORS
`
`21.
`
`The unlawful acts alleged against the Defendants in this class action complaint were
`
`authorized, ordered, or performed by the Defendants’ respective officers, agents, employees,
`
`representatives, or shareholders while actively engaged in the management, direction, or control of the
`
`Defendants’ businesses or affairs. The Defendants’ agents operated under the explicit and apparent
`
`authority of their principals. Each Defendant, and its subsidiaries, affiliates, and agents operated as a
`
`single unified entity.
`22.
`
`Various persons and/or firms not named as Defendants may have participated as co-
`
`conspirators in the violations alleged herein and may have performed acts and made statements in
`
`furtherance thereof. Each acted as the principal, agent, or joint venture of, or for other Defendants with
`
`respect to the acts, violations, and common course of conduct alleged herein.
`V.
`
`FACTUAL ALLEGATIONS
`
`23.
`
`OpenAI creates and sells artificial-intelligence software products. Artificial intelligence is
`
`commonly abbreviated “AI.” AI software is designed to algorithmically simulate human reasoning or
`
`inference, often using statistical methods.
`24.
`
`Certain AI products created and sold by OpenAI are known as large language models. A
`
`large language model (or “LLM” for short) is AI software designed to parse and emit natural language.
`
`Though a large language model is a software program, it is not created the way most software programs
`
`1 2 3 4 5 6 7 8 9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25
`
`26
`
`27
`
`28
`
`
`
`
`
`4
`COMPLAINT
`
`
`
`

`

`Case 4:23-cv-03416-KAW Document 1 Filed 07/07/23 Page 6 of 17
`
`
`
`are—that is, by human software engineers writing code. Rather, a large language model is “trained” by
`
`copying massive amounts of text from various sources and feeding these copies into the model. This
`
`corpus of input material is called the training dataset. During training, the large language model copies
`
`each piece of text in the training dataset and extracts expressive information from it. The large language
`
`model progressively adjusts its output to more closely resemble the sequences of words copied from
`
`the training dataset. Once the large language model has copied and ingested all this text, it is able to
`
`emit convincing simulations of natural written language as it appears in the training dataset.
`25. Much of the material in OpenAI’s training datasets, however, comes from copyrighted
`
`works—including books written by Plaintiffs—that were copied by OpenAI without consent, without
`
`credit, and without compensation.
`26.
`
`Authors, including Plaintiffs, publish books with certain copyright management
`
`information. This information includes the book’s title, the ISBN number or copyright number, the
`
`author’s name, the copyright holder’s name, and terms and conditions of use. Most commonly, this
`
`information is found on the back of the book’s title page and is customarily included in all books,
`
`regardless of genre.
`27.
`
`OpenAI has released a series of large language models, including GPT-1 (released June
`
`2018), GPT-2 (February 2019), GPT-3 (May 2020), GPT-3.5 (March 2022), and most recently GPT-4
`
`(March 2023). “GPT” is an abbreviation for “generative pre-trained transformer,” where pre-trained
`
`refers to the use of textual material for training, generative refers to the model’s ability to emit text, and
`
`transformer refers to the underlying training algorithm. Together, OpenAI’s large language models will
`
`be referred to as the “OpenAI Language Models.”
`28. Many kinds of material have been used to train large language models. Books, however,
`
`have always been a key ingredient in training datasets for large language models because books offer the
`
`best examples of high-quality longform writing.
`29.
`
`For instance, in its June 2018 paper introducing GPT-1 (called “Improving Language
`
`Understanding by Generative Pre-Training”), OpenAI revealed that it trained GPT-1 on BookCorpus,
`
`a collection of “over 7,000 unique unpublished books from a variety of genres including Adventure,
`
`Fantasy, and Romance.” OpenAI confirmed why a dataset of books was so valuable: “Crucially, it
`
`1 2 3 4 5 6 7 8 9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25
`
`26
`
`27
`
`28
`
`
`
`
`
`5
`COMPLAINT
`
`
`
`

`

`Case 4:23-cv-03416-KAW Document 1 Filed 07/07/23 Page 7 of 17
`
`
`
`contains long stretches of contiguous text, which allows the generative model to learn to condition on
`
`long-range information.” Hundreds of large language models have been trained on BookCorpus,
`
`including those made by OpenAI, Google, Amazon, and others.
`30.
`
`BookCorpus, however, is a controversial dataset. It was assembled in 2015 by a team of
`
`AI researchers for the purpose of training language models. They copied the books from a website
`
`called Smashwords that hosts self-published novels, that are available to readers at no cost. Those
`
`novels, however, are largely under copyright. They were copied into the BookCorpus dataset without
`
`consent, credit, or compensation to the authors.
`31.
`
`OpenAI also copied many books while training GPT-3. In the July 2020 paper
`
`introducing GPT-3 (called “Language Models are Few-Shot Learners”), OpenAI disclosed that 15% of
`
`the enormous GPT-3 training dataset came from “two internet-based books corpora” that OpenAI
`
`simply called “Books1” and “Books2”.
`32.
`
`Tellingly, OpenAI has never revealed what books are part of the Books1 and Books2
`
`datasets. Though there are some clues. First, OpenAI admitted these are “internet-based books
`
`corpora.” Second, both Books1 and Books2 are apparently much larger than BookCorpus. Based on
`
`numbers given in OpenAI’s paper about GPT-3, Books1 is apparently about nine times larger; Books2
`
`is about 42 times larger. Since BookCorpus contained about 7,000 titles, this suggests Books1 would
`
`contain about 63,000 titles; Books2 would contain about 294,000 titles.
`33.
`
`But there are only a handful of “internet-based books corpora” that would be able to
`
`deliver this much material.
`34.
`
`As noted in Paragraph 32, supra, the OpenAI Books1 dataset can be estimated to contain
`
`about 63,000 titles. Project Gutenberg is an online archive of e-books whose copyright has expired. In
`
`September 2020, Project Gutenberg claimed to have “over 60,000” titles. Project Gutenberg has long
`
`been popular for training AI systems due to the lack of copyright. In 2018, a team of AI researchers
`
`created the “Standardized Project Gutenberg Corpus,” which contained “more than 50,000 books.”
`
`On information and belief, the OpenAI Books1 dataset is based on either the Standardized Project
`
`Gutenberg Corpus or Project Gutenberg itself, because of the roughly similar sizes of the two datasets.
`
`1 2 3 4 5 6 7 8 9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25
`
`26
`
`27
`
`28
`
`
`
`
`
`6
`COMPLAINT
`
`
`
`

`

`Case 4:23-cv-03416-KAW Document 1 Filed 07/07/23 Page 8 of 17
`
`
`
`35.
`
`As noted in Paragraph 32, supra, the OpenAI Books2 dataset can be estimated to contain
`
`about 294,000 titles. The only “internet-based books corpora” that have ever offered that much
`
`material are notorious “shadow library” websites like Library Genesis (aka LibGen), Z-Library (aka B-
`
`ok), Sci-Hub, and Bibliotik. The books aggregated by these websites have also been available in bulk via
`
`torrent systems. These flagrantly illegal shadow libraries have long been of interest to the AI-training
`
`community: for instance, an AI training dataset published in December 2020 by EleutherAI called
`
`“Books3” includes a recreation of the Bibliotik collection and contains nearly 200,000 books. On
`
`information and belief, the OpenAI Books2 dataset includes books copied from these “shadow
`
`libraries,” because those are the most sources of trainable books most similar in nature and size to
`
`OpenAI’s description of Books2.
`36.
`
`In March 2023, OpenAI’s paper introducing GPT-4 contained no information about its
`
`dataset at all: OpenAI claimed that “[g]iven both the competitive landscape and the safety implications
`
`of large-scale models like GPT-4, this report contains no further details about . . . dataset
`
`construction.” Later in the paper, OpenAI concedes it did “filter[ ] our dataset . . . to specifically
`
`reduce the quantity of inappropriate erotic text content.”
`A.
`
`Interrogating the OpenAI Language Models using ChatGPT
`37.
`
`ChatGPT is a language model created and sold by OpenAI. As its name suggests,
`
`ChatGPT is designed to offer a conversational style of interaction with a user. OpenAI offers ChatGPT
`
`through a web interface to individual users for $20 per month. Through the web interface, users can
`
`choose to use two versions of ChatGPT: one based on the GPT-3.5 model, and one based on the newer
`
`GPT-4 model.
`38.
`
`OpenAI also offers ChatGPT to software developers through an application-
`
`programming interface (or “API”). The API allows developers to write programs that exchange data
`
`with ChatGPT. Access to ChatGPT via the API is billed on the basis of usage.
`39.
`
`Regardless of how accessed—either through the web interface or through the API—
`
`ChatGPT allows users to enter text prompts, which ChatGPT then attempts to respond to in a natural
`
`way, i.e., ChatGPT can generate answers in a coherent and fluent way that closely mimics human
`
`language. If a user prompts ChatGPT with a question, ChatGPT will answer. If a user prompts
`
`
`
`
`
`7
`COMPLAINT
`
`
`
`1 2 3 4 5 6 7 8 9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25
`
`26
`
`27
`
`28
`
`

`

`Case 4:23-cv-03416-KAW Document 1 Filed 07/07/23 Page 9 of 17
`
`
`
`ChatGPT with a command, ChatGPT will obey. If a user prompts ChatGPT to summarize a
`
`copyrighted book, it will do so.
`40.
`
`ChatGPT’s output, like other LLMs, relies on the data upon which it is trained to
`
`generate new content. LLMs generate output based on patterns and connections drawn from the
`
`training data. For example, if an LLM is prompted to generate a writing in the style of a certain author,
`
`the LLM would generate content based on patterns and connections it learned from analysis of that
`
`author’s work within its training data.
`41.
`
`On information and belief, the reason ChatGPT can accurately summarize a certain
`
`copyrighted book is because that book was copied by OpenAI and ingested by the underlying OpenAI
`
`Language Model (either GPT-3.5 or GPT-4) as part of its training data.
`42. When ChatGPT was prompted to summarize books written by each of the Plaintiffs, it
`
`generated very accurate summaries. These summaries are attached as Exhibit B. The summaries get
`
`some details wrong. This is expected, since a large language model mixes together expressive material
`
`derived from many sources. Still, the rest of the summaries are accurate, which means that ChatGPT
`
`retains knowledge of particular works in the training dataset and is able to output similar textual
`
`content. At no point did ChatGPT reproduce any of the copyright management information Plaintiffs
`
`included with their published works.
`VI. CLASS ALLEGATIONS
`
`A.
`
`Class Definition
`43.
`
`Plaintiffs bring this action for damages and injunctive relief as a class action under
`
`Federal Rules of Civil Procedure 23(a), 23(b)(2), and 23(b)(3), on behalf of the following Class:
`
`All persons or entities domiciled in the United States that own a
`United States copyright in any work that was used as training data
`for the OpenAI Language Models during the Class Period.
`
`This Class definition excludes:
`a.
`b.
`c.
`
`any of the Defendants named herein;
`
`any of the Defendants’ co-conspirators;
`
`any of Defendants’ parent companies, subsidiaries, and affiliates;
`
`8
`COMPLAINT
`
`
`
`44.
`
`
`
`
`
`1 2 3 4 5 6 7 8 9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25
`
`26
`
`27
`
`28
`
`

`

`Case 4:23-cv-03416-KAW Document 1 Filed 07/07/23 Page 10 of 17
`
`
`
`B.
`
`d.
`
`e.
`f.
`
`any of Defendants’ officers, directors, management, employees, subsidiaries,
`
`affiliates, or agents;
`
`all governmental entities; and
`
`the judges and chambers staff in this case, as well as any members of their
`
`immediate families.
`
`Numerosity
`45.
`
`Plaintiffs do not know the exact number of members in the Class. This information is in
`
`the exclusive control of Defendants. On information and belief, there are at least thousands of members
`
`in the Class geographically dispersed throughout the United States. Therefore, joinder of all members
`
`of the Class in the prosecution of this action is impracticable.
`C.
`
`Typicality
`46.
`
`Plaintiffs’ claims are typical of the claims of other members of the Class because
`
`Plaintiffs and all members of the Class were damaged by the same wrongful conduct of Defendants as
`
`alleged herein, and the relief sought herein is common to all members of the Class.
`D.
`
`Adequacy
`47.
`
`Plaintiffs will fairly and adequately represent the interests of the members of the Class
`
`because the Plaintiffs have experienced the same harms as the members of the Class and have no
`
`conflicts with any other members of the Class. Furthermore, Plaintiffs have retained sophisticated and
`
`competent counsel who are experienced in prosecuting federal and state class actions, as well as other
`
`complex litigation.
`E.
`
`Commonality and Predominance
`48. Numerous questions of law or fact common to each Class arise from Defendants’
`
`conduct:
`
`
`
`
`
`a. whether Defendants violated the copyrights of Plaintiffs and the Class when they
`
`downloaded copies of Plaintiffs’ copyrighted books and used them to train ChatGPT;
`b. whether ChatGPT itself is an infringing derivative work based on Plaintiffs’ copyrighted
`
`books;
`
`9
`COMPLAINT
`
`
`
`1 2 3 4 5 6 7 8 9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25
`
`26
`
`27
`
`28
`
`

`

`Case 4:23-cv-03416-KAW Document 1 Filed 07/07/23 Page 11 of 17
`
`
`
`c. whether the text outputs of ChatGPT are infringing derivative works based on Plaintiffs’
`
`copyrighted books;
`d. whether Defendants violated the DMCA by removing copyright-management
`
`information (CMI) from Plaintiffs’ copyrighted books.
`e. Whether Defendants were unjustly enriched by the unlawful conduct alleged herein.
`f. Whether Defendants’ conduct alleged herein constitutes Unfair Competition under
`
`California Business and Professions Code section 17200 et seq.
`g. Whether this Court should enjoin Defendants from engaging in the unlawful conduct
`
`alleged herein. And what the scope of that injunction would be.
`h. Whether any affirmative defense excuses Defendants’ conduct.
`i. Whether any statutes of limitation limits Plaintiffs’ and the Class’s potential for recovery.
`49.
`
`These and other questions of law and fact are common to the Class predominate over
`
`any questions affecting the members of the Class individually.
`F.
`
`Other Class Considerations
`50. Defendants have acted on grounds generally applicable to the Class. This class action is
`
`superior to alternatives, if any, for the fair and efficient adjudication of this controversy. Prosecuting the
`
`claims pleaded herein as a class action will eliminate the possibility of repetitive litigation. There will be
`
`no material difficulty in the management of this action as a class action. Further, final injunctive relief is
`
`appropriate with respect to the Class as a whole.
`51.
`
`The prosecution of separate actions by individual Class members would create the risk
`
`of inconsistent or varying adjudications, establishing incompatible standards of conduct for
`
`Defendants.
`
`52.
`53.
`
`VII. CLAIMS FOR RELIEF
`COUNT I
`Direct Copyright Infringement
`17 U.S.C. § 106
`On Behalf of Plaintiffs and the Class
`Plaintiffs incorporate by reference the preceding factual allegations.
`
`As the owners of the registered copyrights in books used to train the OpenAI Language
`
`Models, Plaintiffs hold the exclusive rights to those texts under 17 U.S.C. § 106.
`
`
`
`
`
`10
`COMPLAINT
`
`
`
`1 2 3 4 5 6 7 8 9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25
`
`26
`
`27
`
`28
`
`

`

`Case 4:23-cv-03416-KAW Document 1 Filed 07/07/23 Page 12 of 17
`
`
`
`54.
`
`Plaintiffs never authorized OpenAI to make copies of their books, make derivative
`
`works, publicly display copies (or derivative works), or distribute copies (or derivative works). All those
`
`rights belong exclusively to Plaintiffs under copyright law.
`55.
`
`On information and belief, to train the OpenAI Language Models, OpenAI relied on
`
`harvesting mass quantities of textual material from the public internet, including Plaintiffs’ books,
`
`which are available in digital formats.
`56.
`
`OpenAI made copies of Plaintiffs’ books during the training process of the OpenAI
`
`Language Models without Plaintiffs’ permission. Specifically, OpenAI copied at least Plaintiff
`
`Silverman’s book The Bedwetter; Plaintiff Golden’s book Ararat; and Plaintiff Kadrey’s book Sandman
`
`Slime. Together, these books are referred to as the Infringed Works.
`57.
`
`Because the OpenAI Language Models cannot function without the expressive
`
`information extracted from Plaintiffs’ works (and others) and retained inside them, the OpenAI
`
`Language Models are themselves infringing derivative works, made without Plaintiffs’ permission and
`
`in violation of their exclusive rights under the Copyright Act.
`58.
`
`Plaintiffs have been injured by OpenAI’s acts of direct copyright infringement. Plaintiffs
`
`are entitled to statutory damages, actual damages, restitution of profits, and other remedies provided
`
`by law.
`
`59.
`60.
`
`COUNT 2
`Vicarious Copyright Infringement
`17 U.S.C. § 106
`On Behalf of Plaintiffs and the Class
`Plaintiffs incorporate by reference the preceding factual allegations.
`
`Because the output of the OpenAI Language Models is based on expressive information
`
`extracted from Plaintiffs’ works (and others), every output of the OpenAI Language Models is an
`
`infringing derivative work, made without Plaintiffs’ permission and in violation of their exclusive rights
`
`under the Copyright Act.
`61.
`
`OpenAI has the right and ability to control the output of the OpenAI Language Models.
`
`OpenAI has benefited financially from the infringing output of the OpenAI Language Models.
`
`1 2 3 4 5 6 7 8 9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25
`
`26
`
`27
`
`28
`
`
`
`
`
`11
`COMPLAINT
`
`
`
`

`

`Case 4:23-cv-03416-KAW Document 1 Filed 07/07/23 Page 13 of 17
`
`
`
`Therefore, every output from the OpenAI Language Models constitutes an act of vicarious copyright
`
`infringement.
`62.
`
`Plaintiffs have been injured by OpenAI’s acts of vicarious copyright infringement.
`
`Plaintiffs are entitled to statutory damages, actual damages, restitution of profits, and other remedies
`
`provided by law.
`
`COUNT 3
`Digital Millennium Copyright Act—
`Removal of Copyright Management Information
`17 U.S.C. § 1202(b)
`On Behalf of Plaintiffs and the Class
`
`Plaintiffs incorporate by reference the preceding factual allegations.
`
`Plaintiffs included one or more forms of copyright-management information (“CMI”)
`
`63.
`64.
`
`in each of the Plaintiffs’ Infringed Works, including: copyright notice, title and other identifying
`
`information, the name or other identifying information about the owners of each book, terms and
`
`conditions of use, and identifying numbers or symbols referring to CMI.
`65. Without the authority of Plaintiffs and the Class, OpenAI copied the Plaintiffs’
`
`Infringed Works and used them as training data for the OpenAI Language Models. By design, the
`
`training process does not preserve any CMI. Therefore, OpenAI intentionally removed CMI from the
`
`Plaintiffs’ Infringed Works in violation of 17 U.S.C. § 1202(b)(1).
`66. Without the authority of Plaintiffs and the Class, Defendants created derivative works
`
`based on Plaintiffs’ Infringed Works. By distributing these works without their CMI, OpenAI violated
`
`17 U.S.C. § 1202(b)(3).
`67.
`
`OpenAI knew or had reasonable grounds to know that this removal of CMI would
`
`facilitate copyright infringement by concealing the fact that every output from the OpenAI Language
`
`Models is an infringing derivative work, synthesized entirely from expressive information found in the
`
`training data.
`68.
`
`Plaintiffs have been injured by OpenAI’s removal of CMI. Plaintiffs are entitled to
`
`statutory damages, actual damages, restitution of profits, and other remedies provided by law.
`
`
`
`
`
`
`
`12
`COMPLAINT
`
`
`
`1 2 3 4 5 6 7 8 9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25
`
`26
`
`27
`
`28
`
`

`

`Case 4:23-cv-03416-KAW Document 1 Filed 07/07/23 Page 14 of 17
`
`
`
`COUNT 4
`Unfair Competition
`Cal. Bus. & Prof. Code §§ 17200, et seq.
`On Behalf of Plaintiffs and the Class
`
`69.
`Plaintiffs incorporate by reference the preceding factual allegations.
`70. Defendants have engaged in unlawful business practices, including violating Plaintiffs’
`
`rights under the DMCA, and using Plaintiffs’ Infringed Works to train ChatGPT without Plaintiffs’ or
`
`the Class’s authorization.
`71.
`
`The unlawful business practices described herein violate California Business and
`
`Professions Code section 17200 et seq. (the “UCL”) because that conduct is otherwise unlawful by
`
`violating the DMCA.
`72.
`
`The unlawful business practices described herein violate the UCL because they are
`
`unfair, immoral, unethical, oppressive, unscrupulous or injurious to consumers, because, among other
`
`reasons, Defendants used Plaintiffs’ protected works to train ChatGPT for Defendants’ own
`
`commercial profit without Plaintiffs’ and the Class’s authorization. Defendants further knowingly
`
`designed ChatGPT to output portions or summaries of Plaintiffs’ copyrighted works without
`
`attribution, and they unfairly profit from and take credit for developing a commercial product based on
`
`unattributed reproductions of those stolen writing and ideas.
`73.
`
`The unlawful business practices described herein violate the UCL because consumers
`
`are likely to be deceived. Defendants knowingly and secretively trained ChatGPT on unauthorized
`
`copies of Plaintiffs’ copyright-protected work. Further Defendants deceptively designed ChatGPT to
`
`output without any CMI or other credit to Plaintiffs and Class members whose Infringed Works
`
`comprise ChatGPT’s training dataset. Defendants deceptively marketed their product in a manner that
`
`fails to attribute the success of their product to the copyright-protected work on which it is based.
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`13
`COMPLAINT
`
`
`
`1 2 3 4 5 6 7 8 9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25
`
`26
`
`27
`
`28
`
`

`

`Case 4:23-cv-03416-KAW