GitHub Copilot’s Suggestions Match Snippets of Existing Code Merely 1% of the Time

ARGUMENT

II. PLAINTIFFS FAIL TO STATE A DMCA CLAIM.

A. Plaintiffs Cannot Allege The Required Likelihood Of Infringement.

All of the § 1202 claims fail because Plaintiffs do not allege any likelihood that Copilot will result in copyright infringement of any of their works, as required to state a claim under Stevens v. Corelogic, Inc., 899 F.3d 666, 673 (9th Cir. 2018). To plead a claim under § 1202(a) pertaining to false CMI, a plaintiff must allege that the defendant acted “with intent to induce, enable, facilitate, or conceal infringement.” A claim under § 1202(b) regarding altered or removed CMI similarly requires that a defendant either “have knowledge” or have “reasonable grounds to know” of the same result. Of necessity, then, the tampering with CMI contemplated by § 1202 must be part of a scheme involving purposeful infringement.

That line is critical in a case like this one, where a plaintiff attacks a dual-use technology—a system that is not aimed at infringement, but may nevertheless touch or concern CMI in some way. Whether or not they advance copyright infringement claims, Plaintiffs must still explain in the context of their § 1202 claims how infringement is occurring or is likely to occur. They cannot. At most, Plaintiffs allege Copilot’s suggestions will match snippets of existing code merely 1% of the time. Compl. ¶ 90. And even within that miniscule percentage, they do not try to allege copyright infringement—a virtual impossibility anyway given the various copyright-based obstacles (originality, fair use, etc.) that would preclude a claim. Plaintiffs do not allege a likelihood of infringement, let alone purposeful infringement.

Stevens addressed this scenario in a case involving § 1202(b)’s “reasonable grounds to know” requirement—the lowest of § 1202’s scienter requirements. Photographers challenged the defendant’s software program, which “downsampled” (i.e., condensed the size of) images for use in electronic real estate listings, and in the process failed to preserve the CMI metadata from the plaintiffs’ photographs. 899 F.3d at 671. The plaintiffs nonetheless argued that they satisfied the “reasonable grounds to know” requirement because when CMI is removed, “one method of identifying an infringing photograph has been impaired, [such that] someone might be able to use their photographs undetected.” Id. at 673.

The Ninth Circuit rejected the argument. Although a § 1202 plaintiff “need not show that any specific infringement has already occurred,” it still “must make an affirmative showing … that the defendant was aware or had reasonable grounds to be aware of the probable future impact of its actions.” Id. at 674. And this requires a showing “from which one can infer that future infringement is likely … to occur as a result of the removal or alteration of CMI.” Id. at 675 (emphasis added). The rule makes good sense: If infringement is merely a “general possibility,” id. at 673, there is no plausible inference that the defendant’s conduct was done with the requisite mental state. See also Kelly v. Arriba Soft Corp., 77 F. Supp. 2d 1116, 1122 (C.D. Cal. 1999) (rejecting DMCA claim based on “unintended side effect” of web crawler’s operation), rev’d on other grounds, 336 F.3d 811 (9th Cir. 2003). Courts have dismissed § 1202 claims at the pleading stage for failing to satisfy this requirement. See Harrington v. Pinterest, Inc., No. 20-cv05290, 2021 WL 4033031, at *6 (N.D. Cal. Sept. 3, 2021); Philpot v. Alternet Media, Inc., No. 18-cv-4479, 2018 WL 6267876, at *5 (N.D. Cal. Nov. 30, 2018).

Much as in Stevens, Plaintiffs here challenge technology whose purpose is unrelated to infringement or CMI, but that incidentally may not preserve CMI. Plaintiffs allege that Copilot “has not been trained to provide Attribution,” Compl. ¶ 56, and that, “[o]n information and belief, Defendants could have trained Copilot to include attribution, copyright notices, and license terms.” Compl. ¶ 161. Plaintiffs thus allege that Copilot is CMI-agnostic. Affirmative steps would have to be taken to modify the technology to account for CMI. That is the precise scenario in which Stevens demands a showing that the technology’s incidental effects with respect to CMI have a likelihood of causing, facilitating, or concealing infringement.

To overcome Stevens, Plaintiffs would need to allege facts showing that the technology at issue is likely to result in widespread infringement. Not only do Plaintiffs fail to allege such facts, but it is obvious from the Complaint that Copilot is not such a machine. Plaintiffs allege that suggestions match training data very rarely. Even assuming that were infringement, there is no obligation in copyright law to modify technology to avoid all possibility of infringement. See Sony Corp. of Am. v. Universal City Studios, Inc., 464 U.S. 417 (1984). Finally, and most importantly, it makes no sense to assume that any matching Outputs that might occur constitute copyright infringement. Even when a snippet matches training data, there are a host of additional hurdles to the notion that any particular Output would result in copyright infringement: the copyrightability of such material in light of doctrines of merger and scènes à faire, see Apple Computer, Inc. v. Microsoft Corp., 35 F.3d 1435, 1444-45 (9th Cir. 1994); the requirement of “virtually identical” copying of thinly protected elements, see id. at 1444; the scope of authorization through licensing, compare Sun Microsystems, Inc. v. Microsoft Corp., 188 F.3d 1115 (9th Cir. 1999), with Jacobsen v. Katzer, 535 F.3d 1373 (Fed. Cir. 2008); and fair use, see Google LLC v. Oracle Am., Inc., 141 S. Ct. 1183 (2021). Given these hurdles, Copilot is simply not the type of technology that could ever satisfy the requirements of Stevens.

Similarly, just as in Stevens, Plaintiffs fail to offer any “specific [allegations] that removal of CMI … will impair their policing of infringement.” 899 F.3d at 675. They do not, for example, “aver” that they “have ever used CMI … to prevent or detect copyright infringement.” Id. And just as “a party intent on using a copyrighted photograph undetected can itself remove any CMI metadata,” id. at 676, a party intent on using copyrighted code without reproducing licensing information could readily do so. So, “one cannot plausibly say that removal [by Copilot] ‘will’” make it easier to infringe Plaintiffs’ works. Id.

It is Plaintiffs’ prerogative to not assert a claim for copyright infringement. But Stevens requires plausible allegations that Copilot’s operation with respect to CMI is likely to have that result. Plaintiffs make no such allegations, and the nature of Copilot precludes them from doing so. Count I should be dismissed with prejudice.