Case 1:24-cv-01515-JSR Document 79 Filed 05/16/24 Page 1 of 15
`
`UNITED STATES DISTRICT COURT
`SOUTHERN DISTRICT OF NEW YORK
`
`
`
`
`THE INTERCEPT MEDIA, INC.,
`
`
`Plaintiff,
`
`
`
`v.
`
`
`
`
`
`OPENAI, INC., OPENAI GP, LLC, OPENAI,
`LLC, OPENAI OPCO LLC, OPENAI GLOBAL
`LLC, OAI CORPORATION, LLC, OPENAI
`HOLDINGS, LLC, and MICROSOFT
`CORPORATION,
`
`Case No. 1:24-cv-01515-JSR
`
`
`
`
`
`Defendants.
`
`
`
`
`
`
`DEFENDANT MICROSOFT CORPORATION’S REPLY MEMORANDUM
`IN SUPPORT OF MOTION TO DISMISS THE COMPLAINT
`
`
`
`
`
`
`
`
`
`
`
`
`May 16, 2024
`
`
`
`

`

`Case 1:24-cv-01515-JSR Document 79 Filed 05/16/24 Page 2 of 15
`
`TABLE OF CONTENTS
`
`TABLE OF AUTHORITIES .......................................................................................................... ii
`
`PRELIMINARY STATEMENT .................................................................................................... 1
`
`ARGUMENT .................................................................................................................................. 2
`
`I. THE INTERCEPT LACKS STANDING BECAUSE IT FAILS TO ALLEGE AN ACTUAL
`OR THREATENED INJURY. ................................................................................................. 2
`
`II. THE INTERCEPT FAILS TO STATE A DMCA § 1202 CLAIM.......................................... 5
`
`A. The Complaint Does Not Plausibly Allege “Removal” Of CMI From Copies Of Works. . 5
`
`B. The Complaint Does Not (And Cannot) Plausibly Allege That Microsoft Removed CMI
`Or Distributed Works Lacking CMI. .................................................................................. 7
`
`C. The Complaint Does Not Plausibly Allege A Likelihood That Removal Of CMI During
`Training Of An AI Tool Will Induce, Enable, Facilitate, Or Conceal Infringement. ......... 8
`
`CONCLUSION ............................................................................................................................. 10
`
`
`
`
`
`
`
`
`
`i
`
`

`

`Case 1:24-cv-01515-JSR Document 79 Filed 05/16/24 Page 3 of 15
`
`TABLE OF AUTHORITIES
`
`
`
`Page(s)
`
`Cases
`
`Andersen v. Stability AI Ltd.,
`No. 23-cv-00201-WHO, 2023 WL 7132064 (N.D. Cal. Oct. 30, 2023) ................................................ 6
`
`Ariel (UK) Ltd. v. Reuters Grp. PLC,
`No. 05-cv-9646-JFK, 2006 WL 3161467 (S.D.N.Y. Oct. 31, 2006) ..................................................... 8
`
`Como v. Commerce Oil Co.,
`607 F. Supp. 335 (S.D.N.Y. 1985) ......................................................................................................... 7
`
`Doe 1 v. GitHub, Inc.,
`672 F. Supp. 3d 837 (N.D. Cal. 2023) .................................................................................................... 4
`
`Doe 1 v. GitHub, Inc.,
`No. 22-cv-06823-JST, 2024 WL 235217 (N.D. Cal. Jan. 22, 2024) ...................................................... 6
`
`Free Speech Sys., LLC v. Menzel,
`390 F. Supp. 3d 1162 (N.D. Cal. 2019) .................................................................................................. 6
`
`Hirsch v. CBS Broadcasting Inc.,
`No. 17-cv-1860-PAE, 2017 WL 3393845 (S.D.N.Y. Aug. 4, 2017) ..................................................... 9
`
`Melendez v. Sirius XM Radio, Inc.,
`50 F.4th 294 (2d Cir. 2022) .................................................................................................................... 7
`
`Roberts v. BroadwayHD LLC,
`518 F. Supp. 3d 719 (S.D.N.Y. 2021) .................................................................................................... 3
`
`Spokeo, Inc. v. Robins,
`578 U. S. 330 (2016) ............................................................................................................................. 2
`
`Stevens v. Corelogic, Inc.,
`899 F.3d 666 (9th Cir. 2018) ............................................................................................................ 8, 10
`
`TransUnion LLC v. Ramirez,
`594 U.S. 413 (2021) ........................................................................................................................... 2, 3
`
`Valley Forge Christian Coll. v. Ams. United for Separation of Church and State, Inc.,
`454 U.S. 464 (1982) ............................................................................................................................... 4
`
`Statutes
`
`Copyright Act, 17 U.S.C. §§ 101 et seq.
`
`§ 102 ....................................................................................................................................................... 3
`
`ii
`
`
`
`
`
`

`

`Case 1:24-cv-01515-JSR Document 79 Filed 05/16/24 Page 4 of 15
`Case 1:24-cv-01515-JSR Document 79 Filed 05/16/24 Page 4 of 15
`
`§ 106... eeecsectecssesccsseseescessescssssessssscssssssossssosossssssssesesssssossvssosenssssonavsssssusssssvassessssssnssssssosecsseasoese 3
`§ 106 ....................................................................................................................................................... 3
`
`§ 107... ecssssssssscssevevsscssecsessscessesesessesessessesescsssesasssssssessessssssessecsessesecsssoesessessessusssonsecsasusesonsesenseess 10
`§ 107 ..................................................................................................................................................... 10
`
`§ 1202... esssscscseveevescssecessscesscrsssesssessessesesossecsesossessossesssssssesessususssessecsssasossecsesssossessescusonsones 3, 5, 10
`§ 1202 ........................................................................................................................................... 3, 5, 10
`
`§ 1202(D)..... ee eceescssssecscessescssvescssvscsssnsccsessossssvsssssssesssasoscsssssnssssssssscsseasososassssssssesessasoeseaees 1,5, 8, 10
`§ 1202(b) .................................................................................................................................. 1, 5, 8, 10
`
`iii
`iii
`
`
`
`
`
`
`
`
`
`
`
`
`
`

`

`Case 1:24-cv-01515-JSR Document 79 Filed 05/16/24 Page 5 of 15
`
`PRELIMINARY STATEMENT
`
`
`
`The Intercept’s Opposition confirms that its sparsely pleaded § 1202(b) claims are as
`
`threadbare as they appear.
`
`As to Article III standing, The Intercept concedes that it alleges no injury based on public
`
`dissemination of its works without CMI. Opp. 8. Its asserted harm is nothing more than alleged
`
`removal of attribution information during the entirely internal training and development of an
`
`LLM. And The Intercept also points to no common law recognition that such non-public non-
`
`attribution is a cognizable injury. That ends the inquiry—The Intercept lacks standing. Its only
`
`attempted workaround is highly dubious: It equates the non-attribution in private with copyright
`
`infringement, a claim The Intercept does not bring. The analogy does not work. CMI is merely
`
`information about a copyright-protected work, not copyright-protected material itself. The
`
`Intercept’s claims based on private removal of such information thus could only allege a bare
`
`technical violation of § 1202(b), with no concrete injury required to support standing. Infra § I.
`
`The Intercept also fails to state a claim. Tasked with defending a Complaint light on
`
`factual allegations and heavy on conclusions, The Intercept largely retreats to platitudes about
`
`the plausibility standard. Most of the allegations The Intercept points to are not factual
`
`allegations at all; they are conclusory assertions that cannot support a plausible claim. When
`
`those are cast aside, all that remains of the Complaint are theories, hypotheticals, and rumors
`
`strung together to suggest that (a) because some unidentified Intercept works are allegedly in the
`
`training set, they must have had CMI removed; (b) because Microsoft and OpenAI have a “close
`
`relationship,” Microsoft must have removed CMI from The Intercept’s works; and (c) because
`
`there is some bare possibility that GPT-based products can emit output that matches some text in
`
`the training set, it must be likely that they will do so with The Intercept’s works, and that end-
`
`users will then infringe them further. For The Intercept’s claims to survive, all of these premises
`
`
`
`1
`
`

`

`Case 1:24-cv-01515-JSR Document 79 Filed 05/16/24 Page 6 of 15
`
`must be plausibly alleged. None actually are. The Complaint should be dismissed in its entirety.
`
`Infra § II.
`
`ARGUMENT
`
`I.
`
`THE INTERCEPT LACKS STANDING BECAUSE IT FAILS TO ALLEGE AN
`ACTUAL OR THREATENED INJURY.
`
`Microsoft’s motion explained that the mere removal of CMI in a non-public setting,
`
`without public dissemination of the CMI-stripped copy, causes no concrete injury, and therefore
`
`cannot support standing. Mot. 8-13. In response, The Intercept does not dispute that the
`
`common law recognized no injury based on non-public non-attribution with respect to
`
`intellectual property. Instead, it claims that “the plaintiff gets to pick” its common-law analogy,
`
`Opp. 9 n.3, and selects “copyright infringement,” Opp. 6—the actual invasion of intellectual
`
`property rights themselves. The Intercept thinks its selection solves its non-dissemination
`
`problem because “the common law … recognizes interference with property, without more, as a
`
`concrete injury.” Opp. 7. But the analogy is inapt because CMI is simply not intellectual
`
`property, and The Intercept otherwise alleges no actual or imminent injury to any actual
`
`intellectual property right.
`
`A plaintiff does not get to “pick” whatever common-law analogue seems advantageous.
`
`Opp. 9 n.3. Its “asserted harm” must have a “‘close relationship’ to a harm traditionally
`
`recognized as providing a basis for a lawsuit in American courts.” TransUnion LLC v. Ramirez,
`
`594 U.S. 413, 417 (2021) (quoting Spokeo, Inc. v. Robins, 578 U. S. 330, 340-341 (2016)).
`
`Merely invoking some legally adjacent or similar-sounding claim does not suffice—the injuries
`
`themselves must be analogous in nature. Nor, of course, does it suffice to claim that the
`
`“asserted harm” is a prelude to some other, future harm that may have a common-law analogue.
`
`See id. at 435-36. Thus, the plaintiffs in TransUnion could not prevail by invoking defamation
`
`
`
`2
`
`

`

`Case 1:24-cv-01515-JSR Document 79 Filed 05/16/24 Page 7 of 15
`
`on the theory that the false OFAC labels in the defendant’s files might cause future injury to
`
`reputation if disseminated. Id. at 435; Mot. 8-9.
`
`The Intercept’s attempted analogy to copyright infringement fails these basic tenets. The
`
`Intercept asserts that copyright infringement is analogous to CMI removal because the CMI
`
`removal provisions are housed in “title 17,” i.e., the Copyright Act, and that violations carry
`
`“similar remedies” as copyright infringement. Opp. 6. These observations are irrelevant because
`
`they say nothing about whether the asserted harm—private non-attribution—is analogous to the
`
`invasion of exclusive intellectual property rights.
`
`The Intercept next claims that the CMI-based rights in § 1202 are similar to the
`
`“exclusive rights” in 17 U.S.C. § 106 because “both grant the copyright owner the sole
`
`prerogative to decide how future iterations of the work may differ from the version the owner
`
`published.” Opp. 6-7. This peculiar contrivance manages to obscure the relevant injury for both
`
`traditional copyright infringement and CMI. The harm from copyright infringement is the
`
`invasion of intangible intellectual property rights; the Copyright Act grants “protection” in
`
`“original works of authorship” in defined “categories,” 17 U.S.C. § 102, then confers to the
`
`copyright owner certain “exclusive rights,” id. § 106. Section 1202 in no way expands exclusive
`
`rights in original works of authorship, and many such works have no associated CMI. CMI, by
`
`contrast, is not a copyright-protected “work,” but “information conveyed in connection with
`
`copies … of a work.” Id. § 1202. It is attribution information that “inform[s] the public” of
`
`copyright status. See Roberts v. BroadwayHD LLC, 518 F. Supp. 3d 719, 737 (S.D.N.Y. 2021).
`
`In short, copyright infringement is a property-based injury, and CMI removal is not.
`
`Once this inapt analogy to copyright infringement is dispatched, The Intercept has
`
`nothing left to stand on. Its “asserted harm” is what Microsoft—and The Intercept itself—said it
`
`
`
`3
`
`

`

`Case 1:24-cv-01515-JSR Document 79 Filed 05/16/24 Page 8 of 15
`
`is: an allegation that Defendants “strip away” CMI and therefore do “not … provide attribution”
`
`to “human journalists,” Compl. ¶ 6. The Intercept offers no explanation for how this would
`
`constitute a cognizable injury under the common law given that it is alleged to be occurring out
`
`of public view. See Mot. 10 (explaining that the closest common law analogues—like unfair
`
`competition—all depend on public dissemination). And it expressly disavows “a dissemination-
`
`based injury,” Opp. 8-10, effectively conceding that it cannot bring a suit for damages because it
`
`cannot allege that a GPT-based product has ever disseminated its works without CMI. The
`
`Intercept therefore lacks standing to pursue damages in this case.
`
`
`
`To attempt to salvage some right to remain in court, The Intercept says that “even if
`
`standing did require dissemination, … The Intercept has plausibly alleged facts to support
`
`standing for an injunction.” Opp. 10. The Intercept’s argument depends on the inconceivable
`
`suggestion that a plaintiff need not “plead dissemination of their own works to have standing to
`
`pursue an injunction,” Opp. 10, and that Doe 1 v. GitHub, Inc., 672 F. Supp. 3d 837, 850 (N.D.
`
`Cal. 2023), supports that view. To the contrary, Doe 1 made clear that “the party who invokes
`
`the court’s authority [must] ‘show that he personally has suffered some actual or threatened
`
`injury.’” Id. (quoting Valley Forge Christian Coll. v. Ams. United for Separation of Church and
`
`State, Inc., 454 U.S. 464, 472 (1982)). It then found, in connection with an LLM that
`
`exclusively suggests software code completions, that the plaintiffs had alleged a “substantial
`
`risk” that the LLM “will reproduce Plaintiffs’ licensed code.” Id. at 851 (emphasis added). The
`
`Intercept offers no basis upon which to conclude that the GPT-based products are imminently
`
`likely to reproduce The Intercept’s works without CMI, and indeed offers no response to the
`
`Motion’s argument (at 12) that the Complaint “contains no allegations concerning how any AI
`
`
`
`4
`
`

`

`Case 1:24-cv-01515-JSR Document 79 Filed 05/16/24 Page 9 of 15
`
`product even works” or the “user behavior” that would plausibly elicit a reproduction of The
`
`Intercept’s works. The Intercept therefore lacks standing to pursue injunctive relief as well.
`
`II.
`
`THE INTERCEPT FAILS TO STATE A DMCA § 1202 CLAIM.
`
`The Intercept’s opposition also confirms that its § 1202(b) claims are bare conjecture
`
`based on virtually no plausible factual allegations that Microsoft has done anything to any
`
`Intercept work. Mot. 13-23. The Complaint identifies no works thought to be in the training set
`
`nor offers a basis for concluding that CMI was removed from them. Infra § A. It provides no
`
`plausible basis for the contention that Microsoft created training sets or removed CMI from
`
`works in the training sets (much less from The Intercept’s works). Infra § B. And it proffers no
`
`plausible account of how any alleged removal of CMI from The Intercept’s works will likely
`
`facilitate or conceal piracy, as required for a § 1202(b) claim. Infra § C. Faced with these
`
`shortcomings, The Intercept largely falls back on standards for notice pleading and plausibility
`
`on a motion to dismiss. But those standards are not nearly as forgiving as The Intercept would
`
`need to survive dismissal. The Motion should be granted.
`
`A. The Complaint Does Not Plausibly Allege “Removal” Of CMI From Copies Of
`Works.
`
`As Microsoft’s Motion (at 15-18) explained, The Intercept’s suggestion that CMI has
`
`been removed from its works during training is based purely on unsupported generalities about
`
`how GPT-based products work, without a single example or supporting allegation explaining
`
`why those generalities make its claims plausible. The Intercept does not dispute its failure to
`
`name a copyrighted work it thinks is in a training set, identify any specific CMI removed from it,
`
`or otherwise substantiate its theory that any work contained in the training set must have had
`
`CMI removed. Instead, it asserts that it does not need to do so. This is wrong.
`
`
`
`5
`
`

`

`Case 1:24-cv-01515-JSR Document 79 Filed 05/16/24 Page 10 of 15
`
`In response to Microsoft’s argument that The Intercept fails to identify any works from
`
`which CMI was removed, Mot. 17-18, The Intercept concedes that it “cannot name all its works
`
`contained in [the] training sets,” but says that “[r]equiring a plaintiff to name all its works”
`
`would create perverse incentives to “conceal” CMI violations. Opp. 14. This is distraction,
`
`because The Intercept has not identified a single work from which CMI was allegedly
`
`removed—much less “identical copies” of its works, as is required for a CMI claim, Doe 1 v.
`
`GitHub, Inc., No. 22-cv-06823-JST, 2024 WL 235217, at *8-9 (N.D. Cal. Jan. 22, 2024). The
`
`only allegation in the Complaint is based on some anonymous promise that “thousands” of The
`
`Intercept’s works were included in training sets, without identifying a single one. That is not
`
`enough to survive a motion to dismiss.
`
`The Intercept not only fails to identify the specific works from which CMI was allegedly
`
`removed but also fails to identify the specific CMI that was allegedly removed from those works.
`
`Mot. 17. The Intercept’s response is that Microsoft “ignores the Complaint, which alleges that
`
`Defendants removed author, title, copyright notice, and terms of use from its works,” which “are
`
`all types of CMI.” Opp. 13. Again, this is bare generality. Any plaintiff can recite types of
`
`CMI—that does not mean they have plausibly identified their CMI that was removed from their
`
`works, as both Andersen v. Stability AI Ltd., No. 23-cv-00201-WHO, 2023 WL 7132064, at *11
`
`(N.D. Cal. Oct. 30, 2023) and Free Speech Sys., LLC v. Menzel, 390 F. Supp. 3d 1162, 1175
`
`(N.D. Cal. 2019), confirm.
`
`Most importantly, The Intercept does not meaningfully support the inference that
`
`“everything ever included in the training set … must have been stripped of CMI.” Mot. 15-16.
`
`That is fatal—because The Intercept offers no example of any one of its own works being
`
`stripped of CMI, the plausibility of its claim depends entirely on it establishing that any work
`
`
`
`6
`
`

`

`Case 1:24-cv-01515-JSR Document 79 Filed 05/16/24 Page 11 of 15
`
`contained in the training set will generally be stripped of CMI. Yet The Intercept offers only the
`
`single-sentence assertion that “it is plausible that an LLM doesn’t output CMI precisely because
`
`it wasn’t trained on CMI.” Opp. 17. But as Microsoft explained, Mot. 16, the premises of this
`
`theory are themselves unsupported, so the conclusion is not an available inference. It is just the
`
`sort of conclusory hypothesis the Second Circuit rejected in Melendez v. Sirius XM Radio, Inc.,
`
`50 F.4th 294, 307 (2d Cir. 2022), a case The Intercept does not contend with.
`
`B. The Complaint Does Not (And Cannot) Plausibly Allege That Microsoft
`Removed CMI Or Distributed Works Lacking CMI.
`
`The Intercept also fails to explain why Microsoft’s work with OpenAI is sufficient to
`
`plausibly allege that Microsoft removed CMI or distributed allegedly CMI-less works to
`
`OpenAI. The Intercept readily concedes that the entire basis for its claim that Microsoft
`
`removed The Intercept’s CMI is the “close working relationship” between Microsoft and
`
`OpenAI. Opp. 17; 23. But it offers no authority for the notion that such a relationship is enough
`
`to plausibly allege that anything one entity did, the other must have done also. That is because
`
`“[t]he fact that two companies … do work together” is insufficient “to attribute their acts to one
`
`another.” Como v. Commerce Oil Co., 607 F. Supp. 335, 340 (S.D.N.Y. 1985). Indeed, the mere
`
`fact of a “working relationship” is insufficient to support an inference that Microsoft was even
`
`involved in training OpenAI’s GPT-based products, let alone that it removed CMI from The
`
`Intercept’s works when doing so or shared copies of any such works with OpenAI. Mot. 17-19.
`
`The Intercept’s final attempt to plausibly allege that Microsoft removed CMI from its
`
`works rests on an unsupported (and unsupportable) assertion that Microsoft “ha[s] created
`
`Common Crawl datasets.” Compl. ¶ 31; see Mot. 18. The Intercept clarifies that its theory is
`
`that Microsoft “took data from Common Crawl” and “created training sets from that data,” Opp.
`
`17 n.11, and that it must have removed CMI from Intercept works in that process. Compl. ¶ 38.
`
`
`
`7
`
`

`

`Case 1:24-cv-01515-JSR Document 79 Filed 05/16/24 Page 12 of 15
`
`The Opposition confirms that this is all mere conjecture, offering no response to the Motion’s
`
`direct assertion that The Intercept offers “no basis for [its] information and belief, nor details of
`
`any inquiry it conducted to form such a belief,” Mot. 18.
`
`Not only does the Opposition point to nothing in the Complaint supporting the notion that
`
`Microsoft was involved in training, it points to material from outside the Complaint that
`
`undermines The Intercept’s Common-Crawl theory. Perhaps appreciating the deficiencies in its
`
`Complaint, The Intercept tries to backfill by claiming that “OpenAI has ‘published a list of the
`
`top 1,000 domains present in WebText and their frequency,’” which list has “6,484 (unidentified
`
`URLs from The Intercept’s web domain.” Opp. 14 & nn.7, 9. Leaving aside the impropriety of
`
`adding new allegations in an opposition, Ariel (UK) Ltd. v. Reuters Grp. PLC, No. 05-cv-9646-
`
`JFK, 2006 WL 3161467, at *7 (S.D.N.Y. Oct. 31, 2006), these new allegations suggest that if
`
`The Intercept’s works are in a training set, it is a WebText training set, not a Common Crawl
`
`training set. And the Complaint does not allege, even on a weak information-and-belief basis,
`
`that Microsoft has any connection to these training sets. See Compl. ¶ 31 (distinguishing
`
`between the WebText training sets, which were not created by Microsoft, and the Common
`
`Crawl set, which The Intercept tries to allege Microsoft helped create). The Intercept’s misfire
`
`further confirms that The Intercept cannot tie Microsoft to CMI removal and that its CMI claims
`
`against Microsoft should be dismissed.
`
`C. The Complaint Does Not Plausibly Allege A Likelihood That Removal Of CMI
`During Training Of An AI Tool Will Induce, Enable, Facilitate, Or Conceal
`Infringement.
`
`Finally, The Intercept fails to allege a likelihood of future infringement, as necessary to
`
`establish Microsoft’s awareness that removal of CMI from a work in a training set will somehow
`
`“induce, enable, facilitate, or conceal” such infringement under 17 U.S.C. § 1202(b). Mot. 20-
`
`23. The Opposition does not dispute that this element of the claim, though “ultimately a scienter
`
`
`
`8
`
`

`

`Case 1:24-cv-01515-JSR Document 79 Filed 05/16/24 Page 13 of 15
`
`requirement,” requires allegations that “infringement is objectively ‘likely.’” Mot. 21 (quoting
`
`Stevens v. Corelogic, Inc., 899 F.3d 666, 675 (9th Cir. 2018)). But it points to nothing in the
`
`Complaint plausibly establishing that GPT users, in the real world, are likely to infringe The
`
`Intercept’s works as a result of the absence of CMI—another fatal failing in light of The
`
`Intercept’s inability to point to a single example of its works being output at all.
`
`The Intercept begins by stressing that scienter allegations may be “sparse” and that courts
`
`in the Second Circuit are “lenient” when it comes to pleading them. Opp. 18. But even in the
`
`scienter case The Intercept likes best, Hirsch v. CBS Broadcasting Inc., the plaintiff identified the
`
`specific work from which he claimed CMI had been removed (a particular photo); described the
`
`features of the CMI that was removed (a gutter credit); and explained how the work, stripped of
`
`CMI, had been disseminated to the public (via a television program). See No. 17-cv-1860-PAE,
`
`2017 WL 3393845, at *8 (S.D.N.Y. Aug. 4, 2017). Only with those allegations describing the
`
`actual dissemination of a photo lacking a gutter credit to the public did the court conclude that
`
`the “sparse” scienter allegations sufficed to show awareness of likely future infringement. Id.
`
`As explained (at 5-7), The Intercept offers no such allegations about its works and the CMI that
`
`was allegedly removed; nor does it offer factual matter demonstrating that—or how—its works
`
`would be reproduced to the public without CMI, making it impossible to infer from context that
`
`further infringement resulting from the absence of CMI is likely.
`
`This same problem undermines The Intercept’s reliance on the general allegation that
`
`reproducing works without CMI would “misinform the public about the source of the content
`
`Defendants’ products provide to users.” Opp. at 22. Again, as Microsoft explained in its Motion
`
`(at 15), The Intercept provides not a single example of any such work ever being reproduced, nor
`
`even describes with any detail what such a reproduction would look like in the context of the
`
`
`
`9
`
`

`

`Case 1:24-cv-01515-JSR Document 79 Filed 05/16/24 Page 14 of 15
`
`generative AI tools at issue. Would it be a paraphrase? A few sentences in a larger response?
`
`What user prompt might elicit it? But even leaving that defect aside, The Intercept would still
`
`need an account of why (a) an end-user would be inclined to make a further use of an
`
`undescribed, unattributed natural-language output; (b) that use would be infringing—versus a
`
`fair use, 17 U.S.C. § 107; and (c) why a user so inclined would not make such a use if it
`
`observed CMI. That is precisely the teaching of Stevens v. Corelogic, as well as the several cases
`
`that have applied it in dismissing § 1202 claims at the pleading stage. Mot. 23 (collecting cases).
`
`The Intercept has no answer. It acknowledges that the only allegations in the Complaint
`
`are generalizations suggesting that users might “morally oppose stealing others’ works or fear
`
`liability.” Opp. 23. None of this is grounded in actual, observed behavior of real people using
`
`GPT-based products. And so The Intercept falls back on the generic statement that “one function
`
`of CMI is to ‘discourage piracy,’” and argues that because “Congress concluded that CMI will
`
`discourage infringement, that is surely enough to survive a motion to dismiss.” Opp. 23. This
`
`circular reasoning fails—the whole point of requiring a showing of objectively likely
`
`infringement is that the removal of CMI may be so incidental or so unlikely to facilitate
`
`infringement that § 1202 liability would not serve the aim of thwarting piracy. The need for such
`
`a limitation is most pressing in cases like this one, involving not purposeful distribution of
`
`pirated copies, but multi-purpose technology that end-users put to countless non-infringing uses.
`
`Because The Intercept has not plausibly alleged that these end-users are likely to infringe The
`
`Intercept’s works as a result of absent CMI, the § 1202(b) claims should be dismissed.
`
`The Court should grant the motion to dismiss in its entirety.
`
`CONCLUSION
`
`
`
`
`
`10
`
`

`

`Case 1:24-cv-01515-JSR Document 79 Filed 05/16/24 Page 15 of 15
`
`Orrick, Herrington & Sutcliffe LLP
`Dated: May 16, 2024
`
`
`
`San Francisco, CA
`
`
` By:
`Annette L. Hurst (admitted pro hac vice)
`
`The Orrick Building
`405 Howard Street
`San Francisco, CA 94105-2669
`Telephone: (415) 773-5700
`ahurst@orrick.com
`
`Lisa T. Simpson
`51 West 52nd Street
`New York, NY 10019
`Telephone: (212) 506-3767
`lsimpson@orrick.com
`
`Christopher J. Cariello
`51 West 52nd Street
`New York, NY 10019
`Telephone: (212) 506-3778
`ccariello@orrick.com
`
`Attorneys for Defendant
`Microsoft Corporation
`
`
`
`
`
`11
`
`/s/ Annette L. Hurst
`
`