The Volokh Conspiracy
Mostly law professors | Sometimes contrarian | Often libertarian | Always independent
California Wants to Hold Social Media Platforms Liable for User Posts Containing Bias- and Political-Hostility-Motivated Threats
California SB 771, which is now on Governor Newsom's desk for signature, would add a new statute that provides the following (some structure added):
(a) A social media platform [that has >$100M in annual revenues] that
- violates [California Civil Code] Section 51.7, 51.9, 52, or 52.1 through its algorithms that relay content to users
- or aids, abets, acts in concert, or conspires in a violation of any of those sections,
- or is a joint tortfeasor in a violation of any of those sections,
shall … be liable to a prevailing plaintiff for a civil penalty for each violation sufficient to deter future violations but not to exceed [$1M for knowing violations, and $500K for reckless violations, potentially doubled if the platform knew, or should have known, that the plaintiff was a minor].
(b) (1) For purposes of this section, deploying an algorithm that relays content to users may be considered to be an act of the platform independent from the message of the content relayed.
(2) A platform shall be deemed to have actual knowledge of the operations of its own algorithms, including how and under what circumstances its algorithms deliver content to some users but not to others.
To explain (with some oversimplification) the statutory cross-references,
- Section 51.7 bans violence and threats of violence based on actual or perceived political affiliation, position in a labor dispute, or race, religion, immigration status, etc.
- Section 51.9 bans sexual harassment in a wide range of business relationships.
- Section 52 imposes liability for violations of 51.7, of 51.9, and of California bans on discrimination in places of public accommodation, and discrimination and boycotts by businesses.
- Section 52.1 bans interfering ("whether or not under color of law") by threat, intimidation, or coercion with the exercise of any constitutional rights (including free speech rights).
The legislature's background findings, from section 1 of SB 771, seem to suggest the legislature is concerned specifically about "targeted threats, violence, and coercive harassment, particularly when directed at historically marginalized groups," "especially … in light of rising incidents of hate-motivated harm, as documented across the state":
- [H]ate crimes involving anti-immigrant slurs increased by 31 percent ….
- [A] 400-percent rise in anti-LGBTQ+ disinformation and harmful rhetoric on major social media platforms.
- [A]nti-Jewish bias events rose by 52.9 percent and anti-Islamic bias events rose by 62 percent in 2023.
- Paid advertisements promoting violence against women, including language calling for beatings and killings, [have been] successfully placed and distributed on major social media platforms.
The legislature adds that "[t]he purpose of this act is not to regulate speech or viewpoint but to clarify that social media platforms, like all other businesses, may not knowingly use their systems to promote, facilitate, or contribute to conduct that violates state civil rights laws."
Now the law of course already bans aiding and abetting criminal or tortious behavior. But, as the Supreme Court concluded with regard to federal law in Twitter, Inc. v. Taamneh (2023), such liability generally requires some special steps on the defendant's part to aid the illegal actions. In particular, the Court rejected an aiding and abetting claim based on Twitter's knowingly hosting ISIS material and its algorithm supposedly promoting it, because Twitter didn't give ISIS any special treatment:
- "ISIS was able to upload content to the platforms and connect with third parties, just like everyone else."
- "[D]efendants' recommendation algorithms matched ISIS-related content to users most likely to be interested in that content—again, just like any other content."
- "All the content on [the] platforms is filtered through these algorithms, which allegedly sort the content by information and inputs provided by users and found in the content itself. As presented here, the algorithms appear agnostic as to the nature of the content, matching any content (including ISIS' content) with any user who is more likely to view that content."
- "[T]here are no allegations that defendants treated ISIS any differently from anyone else."
But the new California law seems to intentionally set forth aiding-and-abetting liability under California law that goes well beyond what Taamneh recognized under federal law. Coupled with the new statute's subsection (b)(2)—"A platform shall be deemed to have actual knowledge of the operations of its own algorithms, including how and under what circumstances its algorithms deliver content to some users but not to others"—the knowledge element required under the existing California tort law of aiding and abetting will often be satisfied.
Say a platform's algorithm delivers content to users that contains threats that are based on political affiliation, race, religion, sexual orientation, etc., just because users have shown an interest in the content (not because of any purposeful desire to promote such threatening content in general). The platform may be liable, on the theory that it is "deemed to have actual knowledge" of what its algorithms do. Likewise if the posts contain threats aimed at interfering with free speech, free exercise of religion, and other rights. And of course if platforms are required (on pain of liability) to take down illegal threats, they will likely also take down other material that they're worried might be seen as threatening by a future plaintiff, judge, and jury.
I'm pretty sure that such liability will be precluded by 47 U.S.C. § 230. Courts have held that, under § 230, online providers are immune from liability for speech posted by their users, even under an aiding and abetting theory, unless they deliberately craft their sites to help promote illegal conduct. Here's an example, from Wozniak v. YouTube, LLC (Cal. App. 2024) (yes, Apple co-founder Steve Wozniak):
Here, plaintiffs have not alleged that defendants undertook any … acts to actively and specifically aid the illegal behavior. Instead, they allege only that YouTube's neutral algorithm results in recommending the scam videos to certain targeted users. For instance, the [Complaint] alleges that "YouTube's state-of-the-art algorithm tailors its recommended videos to its users based on a variety of personal information and data that YOUTUBE and GOOGLE collect about their users, including 'clicks, watch time, likes/dislikes, comments, freshness, and upload frequency.'" There is no allegation that YouTube has done anything more than develop and use a content-neutral algorithm.
Courts have consistently held that such neutral tools do not take an interactive computer service outside the scope of section 230 immunity. In Dyroff v. Ultimate Software Group, Inc. (9th Cir. 2019), for instance, the plaintiff was the family of a man who had died after using fentanyl-laced heroin, which he had acquired following communications on defendant's online messaging board. The plaintiff contended the messaging board created content because it "used features and functions, including algorithms, to analyze user posts … and recommend other user groups." The Ninth Circuit rejected the argument, holding that "[t]hese functions—recommendations and notifications—[were] tools meant to facilitate the communication and content of others," and were "not content in and of themselves."
The online message board employed neutral tools similar to the ones challenged by plaintiffs here, and there is no allegation that the algorithms treat the scam content differently than any other third party content. (Ibid.; see also Gonzalez, supra, 2 F.4th at p. 896 ["a website's use of content-neutral algorithms, without more, does not expose it to liability for content posted by a third party"]; Roommates, supra, 521 F.3d at p. 1171 [website not transformed into content creator by virtue of supplying neutral tools that deliver content in response to user inputs]; cf. Liapes, supra, 95 Cal.App.5th at p. 929 [Facebook's tools were not neutral—rather than merely proliferate and disseminate content as a publisher, they created, shaped, and developed content by requiring users to provide information used to contribute to discriminatory unlawfulness].)
The last-cited case, Liapes, helps illustrate where § 230 immunity might not apply: There, Facebook was held potentially liable for discrimination because "It designed and created an advertising system, including the Audience Selection tool, that allowed insurance companies to target their ads based on certain characteristics, such as gender and age." But absent such specific design decisions that promote forbidden discriminatory advertising, § 230 prohibits liability that's based on simply deploying a "neutral algorithm [that] results in recommending [certain material] to certain targeted users" "based on a variety of personal information and data that" the platform collects about users, "including 'clicks, watch time, likes/dislikes, comments, freshness, and upload frequency.'"
So if I'm right, § 230 will preclude California law from imposing such liability on platforms. But it does appear that the California Legislature wants to impose that liability.
