In Re Mosaic LLM Litigation

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
United States District Court
Northern District of California
UNITED STATES DISTRICT COURT
NORTHERN DISTRICT OF CALIFORNIA
In Re Mosaic LLM Litigation Case No. 24-cv-01451-CRB (LJC)
ORDER RESOLVING ECF NO. 176
Re: Dkt. No. 176
Before the Court is the parties’ joint discovery letter brief regarding additional proposed
search terms and Plaintiffs’ allegedly deficient responses to interrogatories. The Court assumes
the parties’ familiarity with the procedural and factual background of the case.
Defendants’ request that Plaintiffs run search terms 1 to 27 is denied. Defendants’ request
that Plaintiffs run search terms 28 to 99 is denied, and in lieu of these search terms, Defendants are
ordered to identify up to ten search terms regarding “licensing agreements for AI training data and
related commentary.” The parties are ordered to further meet and confer to reach an agreement on
the replacement terms. Plaintiffs are ordered to supplement their responses to Interrogatory
No. 11 (No. 14 for Plaintiff Keene).
I. Additional Search Terms
The parties agreed on thirty search terms that Plaintiffs would run against their ESI.
Defendants now request that Plaintiffs run an additional ninety-nine search terms. ECF No. 176
at 3. Defendants’ proposed additional search terms, attached to the instant discovery brief at
Exhibit A, fall into two categories. Terms 1 to 27 relate to third-party AI tools (such as
“ChatGPT!” “Ernie!” “Claude!”) and terms 28 to 99 relate to licensing agreements (such as
“Musk! AND (licens! OR train! OR deal! OR agreem! OR data! OR right! OR AI! OR GenAI!
Case 3:24-cv-01451-CRB Document 180 Filed 11/04/25 Page 1 of 7

2
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
United States District Court
Northern
 District of California
OR pay! OR contract! OR royalt!). See ECF No. 176-1. Plaintiffs object that terms 1 to 27 are
irrelevant and overbroad and that terms 28 to 99 are not proportionate to the needs of the case. See
ECF No. 176 at 4.
A. Terms 1 to 27
Terms 1 to 27 “seek documents regarding Plaintiffs’ use of, and commentary about, third-
party AI tools.” ECF No. 176 at 1. Defendants contend that terms 1 to 27 seek documents
relevant to their fair-use defense, arguing that “Plaintiffs’ use of, and views about, AI tools are
directly relevant to (1) the transformative nature of these technologies, (2) the public benefits they
deliver, and (3) Plaintiffs’ theory that they cause market harm.” ECF No. 176 at 1. They also
argue that the terms are responsive to their RFP Nos. 16 (seeking documents relating to Plaintiffs’
use of LLMs), 24 (seeking documents concerning the market for copyrighted works in training
LLMs or other AI technology) and 32 (seeking documents regarding Plaintiffs’ knowledge of the
use of their copyrighted works for training LLMs). Id.; see, e.g. ECF No. 176-4 at 32-35 (Plaintiff
Rebecca Makkai’s Objections to Defendants’ First Set of RFPs). Plaintiffs argue that their own
use of LLMs is irrelevant to Defendants’ fair use defense and that, in any event, they have already
produced documents hitting on the search string “‘generative ai*’ OR ‘genai*’ OR ‘gen ai*’ OR
‘artificial intelligence’ OR ‘ai*’.” ECF No. 176 at 4. Without conceding relevance, they argue
that this search string has already hit on documents related to their use of third-party LLMs, and
running 27 additional search terms consisting of the names of third-party companies and LLMs
would be redundant.
The Court agrees with Plaintiffs. Federal Rule of Civil Procedure 26(b) provides that
“[p]arties may obtain discovery regarding any nonprivileged matter that is relevant to any party's
claim or defense and proportional to the needs of the case.” Defendants have not shown that their
proposed terms are designed to hit on relevant documents. They contend that terms 1 to 27 will
hit on documents that are “directly relevant to the first and fourth fair-use factors” of their fair-use
defense. The Copyright Act provides that use of a copyrighted work for purposes considered
“fair,” such as for “criticism, comment, news reporting, teaching, … scholarship, or research, is
not an infringement of copyright.” 17 U.S.C. § 107. Although fair use is a flexible concept, the
Case 3:24-cv-01451-CRB Document 180 Filed 11/04/25 Page 2 of 7

3
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
United States District Court
Northern
 District of California
Copyright Act instructs courts to consider four factors to assess if an otherwise infringing use is
fair:
(1) the purpose and character of the use, including whether such use
is of a commercial nature or is for nonprofit educational purposes;
(2) the nature of the copyrighted work; (3) the amount and
substantiality of the portion used in relation to the copyrighted work
as a whole; and (4) the effect of the use upon the potential market for
or value of the copyrighted work.
Id.; Google LLC v. Oracle America, Inc., 593 U.S. 1, 20 (2021). The focus of the first factor is “to
see … whether the new work merely supersede[s] the objects of the original creation or instead
adds something new, with a further purpose or different character, altering the first with new
expression, meaning, or message.” Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569, 579 (1994)
(quotation marks and citations omitted); see Andy Warhol Found. for the Visual Arts, Inc. v.
Goldsmith, 598 U.S. 508, 528 (2023). “[I]t asks, in other words, whether and to what extent the
new work is transformative.” Campbell, 510 U.S. at 579 (quotation marks and citations omitted).
Thus, the relevant inquiry under factor one is whether, and to what extent, MosaicML’s alleged
use of Plaintiffs’ works alters or transforms Plaintiffs’ works. The undersigned does not see, and
Defendants have not explained, how documents regarding Plaintiffs’ use of or commentary about
other generative AI models would bear on “whether and to what extent” MosaicML’s alleged use
of Plaintiffs’ works is transformative. Id.
The fourth factor, considered the most important, “requires courts to consider … the extent
of market harm caused by the particular actions of the alleged infringer” and “whether unrestricted
and widespread conduct of the sort engaged in by the defendant ... would result in a substantially
adverse impact on the potential market for the original.” Id. (quotation marks omitted).
Defendants contend that terms 1 to 27 will allow them to assess “whether any market impact can
be traced to the MPT models, which are not generally available, as distinct from generally
available AI models.” ECF No. 176 at 2. The Court agrees, at a general level, that documents
showing the market harm to Plaintiffs caused by third-party generative AI models could be
relevant both to show the impact of “unrestricted and widespread conduct of the sort engaged in
by the defendant” and to show the impact of Defendants’ versus third parties’ models. Campbell,
Case 3:24-cv-01451-CRB Document 180 Filed 11/04/25 Page 3 of 7

4
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
United States District Court
Northern
 District of California
510 U.S. at 579. But terms 1 to 27 are not tailored to hit on documents regarding the harm caused
by third party models but rather designed to hit on all documents that include the names of certain
models. This will presumably encompass documents regarding Plaintiffs’ “use of” and opinions
of these models, which Defendants have not shown are relevant to the fourth factor. ECF No. 176
at 1. Plaintiffs have already produced documents responsive to the string “‘generative ai*’ OR
‘genai*’ OR ‘gen ai*’ OR ‘artificial intelligence’ OR ‘ai*’.” Absent a more robust showing of
relevance, requiring them to run the names of twenty-seven specific generative AI models and
companies against their ESI is disproportionate to the needs of the case. Defendants’ request that
Plaintiffs run terms 1 to 27 is denied.
B. Terms 28 to 99
Terms 28 to 99 “concern licensing agreements for AI training data and related
commentary.” ECF No. 176 at 2. The terms generally use the same format: the name of a
publisher, technology company or CEO involved in generative AI, or news source, in conjunction
with the string: “(licens! OR train! OR deal! OR agreem! OR data! OR right! OR AI! OR GenAI!
OR pay! OR contract! OR royalt!).” See ECF No. 176-1.1 Although Plaintiffs do not dispute that
documents regarding licensing of their asserted copyrighted works are relevant, they argue that
terms 28 to 99 are overbroad and disproportionate to the needs of the case. They note that they
have already collected and produced documents hitting on the parties’ negotiated “search terms
relating to agreements with publishers and third parties regarding the asserted works,” and have
agreed to run the string “(Licens! OR agree! OR collab! OR partner! OR permiss!) w/10 data!
AND train! OR ‘training data’ OR dataset!)” and produce responsive documents, which they
represent will cover documents “involving seeking or contemplating the use of works as training
date for AI.” ECF No. 176 at 4 n.6. Plaintiffs contend that terms 28 and 99, which are not
“connected to Plaintiffs’ asserted works” or licensing for AI training purposes, are thus “overbroad
and duplicative.” Id. at 4.
1 For example, term 48 is “((Open! /3 AI!) OR OpenAI!) AND (licens! OR train! OR deal! OR
agreem! OR data! OR right! OR AI! OR GenAI! OR pay! OR contract! OR royalt!).” ECF No.
146-1 at 4. Term 92 is “(Le /3 Monde!) AND (licens! OR train! OR deal! OR agreem! OR data!
OR right! OR AI! OR GenAI! OR pay! OR contract! OR royalt!).” Id. at 6.
Case 3:24-cv-01451-CRB Document 180 Filed 11/04/25 Page 4 of 7

5
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
United States District Court
Northern
 District of California
The Court agrees that terms 28 to 99 are unreasonably overbroad. Term 89 is illustrative.
This proposed term is: “Time! AND (licens! OR train! OR deal! OR agreem! OR data! OR right!
OR AI! OR GenAI! OR pay! OR contract! OR royalt!)” and hits on 101,797 unique documents
across the five Plaintiffs. This presumably encompasses documents pertaining to any
“agreement,” any “deal,” or any “license” the Plaintiffs have with Time, in no way limited to
documents regarding Plaintiffs’ asserted works or “licensing agreements for AI training data and
related commentary.” Similarly, term 53—“author! AND (licens! OR train! OR deal! OR agreem!
OR data! OR right! OR AI! OR GenAI! OR pay! OR contract! OR royalt!)”—hitting on 54,431
unique documents, is not tailored to hit on relevant and responsive documents. Requiring
Plaintiffs to run search terms that will likely hit on tens of thousands of unresponsive and
irrelevant documents is overly burdensome and not proportional to the needs of the case. See Fed.
R. Civ. P. 26(b).
Plaintiffs ask the Court to deny Defendants’ request for them to run terms 28 to 99, and
instead permit Plaintiffs “to run the search terms proposed in their July 17, 2025 proposals
identified in Exhibit F.” ECF No. 176 at 5. There is no Exhibit F attached to the instant dispute,
and the Court assumes that Plaintiffs are referring to their proposal at Exhibit E, where they offer
to run the following search terms: “(big! /3 tech!) w/4 (license OR licensing OR licenses OR deal
OR deals OR agreem! OR right OR rights)” and “Transform! w/4 (license OR licensing OR
licenses OR deal OR deals OR agreem! OR right OR rights).” 176-5 at 2. The Court declines to
adopt Plaintiffs proposal and instead orders the following:
First, if Plaintiffs have not already done so, they must run the string “(Licens! OR agree!
OR collab! OR partner! OR permiss!) w/10 data! AND train! OR ‘training data’ OR dataset!)” and
produce responsive documents immediately.
Second, in lieu of terms 28 to 99, by November 6, 2025, Defendants may propose up to ten
replacement search terms designed to hit on documents concerning “licensing agreements for AI
training data and related commentary.” ECF No. 176 at 2. If Plaintiffs object to Defendants’
proposed terms, they must send their objections to Defendants by November 10, 2025, and, for
each term they object to they must (1) provide hit counts (using the same format as the parties’
Case 3:24-cv-01451-CRB Document 180 Filed 11/04/25 Page 5 of 7

6
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
United States District Court
Northern
 District of California
Exhibit A) and (2) identify which portion(s) of the proposed terms generate noise. The parties
must then meet and confer in person or over video conference and engage in a good-faith effort to
reach an agreement as to the ten search terms. If the parties are unable to resolve their dispute,
they may submit a joint status report to the Court, not to exceed five pages, by November 13,
2025. The status report must include Defendants’ proposed terms and Plaintiffs’ objections and
hit reports.
II. Response to Interrogatory No. 11/No. 14
In addition to the search term dispute, Defendants complain that the five Plaintiffs’
responses to Interrogatory No. 11 (Interrogatory No. 14 for Plaintiff Keene) are deficient. ECF
No. 176 at 2. Interrogatory No. 11/14 requests that each Plaintiff “identify all locations on the
internet, including any websites or data repositories, where any copies of Your Asserted Works, in
whole or in part, are or have been available, and for each, state whether the Work was made
available with Your authorization and any steps You have taken to have those copies
removed.” ECF No. 176-4 at 4 (Interrogatories to Plaintiff Rebecca Makkai). Plaintiffs argue that
as the internet includes an estimated 1.2 billion webpages, it is unreasonable for them to identify
all locations where their works are available. ECF No. 176 at 4. They propose identifying the
“traditional distribution channels” where their works are available and any takedown requests they
have made. Id. at 5. Defendants characterize Plaintiffs’ objections as unnecessary hand-waiving
and clarify that Plaintiffs do not need to “search the entire internet,” but rather just respond “to the
fullest extent possible” after making a reasonable investigation. Id.
Federal Rule of Civil Procedure 33(b) requires that interrogatories must be answered
“fully.” “In general, a responding party is not required ‘to conduct extensive research in order to
answer an interrogatory, but a reasonable effort to respond must be made.’” Gorrell v. Sneath,
292 F.R.D. 629, 632 (E.D. Cal. 2013) (quoting Haney v. Saldana, 04-cv-05936, 2010 WL
3341939, at *3 (E.D. Cal. Sept. 21, 2007)). “A party has an obligation to conduct a reasonable
inquiry into the factual basis of its discovery responses.” Nat’l Acad. of Recording Arts & Scis.,
Inc. v. On Point Events, LP, 256 F.R.D. 678, 680 (C.D. Cal. 2009). The Court agrees with
Defendants that, as Plaintiffs need not “conduct extensive research” to respond to the
Case 3:24-cv-01451-CRB Document 180 Filed 11/04/25 Page 6 of 7

7
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
United States District Court
Northern
 District of California
interrogatory, Plaintiffs are not required to “scour all locations on the internet and locate every
single webpage where Plaintiffs’ copyrighted works appear.” Gorrell, 292 F.R.D. at 632
(quotation marks omitted); ECF No. 176 at 5. Plaintiffs may satisfy their obligation to fully
respond by identifying all locations on the internet where they know their works are available,
conduct a reasonable and good faith search to identify other locations, identify the “traditional
distribution channels” containing Plaintiff’s asserted works, and describe all efforts Plaintiffs have
taken to remove unauthorized copies. Plaintiffs must supplement their responses to Interrogatory
Nos. 11/14 by November 12, 2025.
IT IS SO ORDERED.
Dated: November 4, 2025
LISA J. CISNEROS
United States Magistrate Judge
Case 3:24-cv-01451-CRB Document 180 Filed 11/04/25 Page 7 of 7