Moody v. NetChoice, LLC

Moody, Attorney General of Florida, et al. v. NetChoice, LLC, dba NetChoice, et al., No. 22-277, decided July 1, 2024 together with NetChoice, LLC, dba NetChoice, et al. v. Paxton, Attorney General of Texas, No. 22-555, is a Supreme Court decision on facial First Amendment challenges to two state social-media statutes. Justice Kagan delivered the opinion of the Court, which vacated the judgments of the Eleventh and Fifth Circuits and remanded both cases because neither court had conducted a proper analysis of the facial challenges before it (Moody v. NetChoice, LLC — slip opinion (SCOTUS, 2024)).

The decision resolved no state law's constitutionality. Its significance for AI policy lies in what the Court said while explaining the analysis the lower courts should have performed: that a private entity's compilation and curation of third-party speech into an expressive product is itself protected expression. That proposition is the doctrinal hinge on which challenges to AI output-regulation statutes are argued, and it is treated at Algorithmic Speech Doctrine.

Case information

Background

Florida enacted SB 7072 in 2021, regulating "social media platforms" and restricting varied ways of "censor[ing]" or otherwise disfavoring posts based on their content or source. It also mandated that a platform provide an explanation to a user whenever it removes or alters any of her posts, with notice delivered within seven days containing both a "thorough rationale" for the action and an account of how the platform became aware of the targeted material.

Texas enacted HB 20 the same year, prohibiting covered platforms from "censor[ing]" a "user's expression" based on the "viewpoint" it contains, and requiring that "concurrently with the removal" of user content the platform "notify the user" and "explain the reason the content was removed" (Moody v. NetChoice, LLC — slip opinion (SCOTUS, 2024)).

The trade associations brought facial challenges to both statutes. The Eleventh Circuit largely upheld a preliminary injunction against the Florida law; the Fifth Circuit reversed an injunction against the Texas law, producing a circuit split that the Court took up.

Holding

The Court vacated and remanded both judgments. Neither circuit had considered the full range of the statutes' applications, or asked whether a substantial number of those applications were unconstitutional relative to the statutes' plainly legitimate sweep. Both courts had instead focused on particular applications — the curation of Facebook's News Feed, principally — without addressing the many other services and functions the laws reached.

On the merits question underlying the analysis, the opinion stated that "the First Amendment offers protection when an entity engaged in compiling and curating others' speech into an expressive product of its own is directed to accommodate messages it would prefer to exclude." Platforms curating their feeds engage in protected editorial control, and laws forcing them to carry or promote content they regard as objectionable interfere with that protected activity.

On the individualized-explanation provisions, the Court framed the inquiry as whether the required disclosures unduly burden expression, referencing Zauderer v. Office of Disciplinary Counsel. It noted that the Eleventh Circuit had found those provisions "unduly burdensome and likely to chill platforms' protected speech," while the Fifth Circuit had found them not unduly burdensome because platforms needed only to "scale up a 'complaint-and-appeal process' they already used."

Lineup and separate writings

Kagan, J., delivered the opinion of the Court, joined in full by Roberts, C.J., and Sotomayor, Kavanaugh, and Barrett, JJ., and joined by Jackson, J., as to Parts I, II, and III–A. Four separate writings followed, and they carry most of the doctrinal signal on how the Court's members would treat algorithmic ranking as distinct from human curation (Moody v. NetChoice, LLC — slip opinion (SCOTUS, 2024)).

Barrett, concurring. Barrett joined the Court in full but argued that an as-applied challenge would have served the platforms better. Her concurrence distinguishes an algorithm built to implement a human editorial decision — protected, "even if the algorithm does most of the deleting without a person in the loop" — from one that "just presents automatically to each user whatever the algorithm thinks the user will like," for which "[t]he First Amendment implications of the Florida and Texas laws might be different." She extends the question to models directly: "If the AI relies on large language models to determine what is 'hateful' and should be removed, has a human being with First Amendment rights made an inherently expressive 'choice . . . not to propound a particular point of view'?" She concludes that "technology may attenuate the connection" between a content-moderation action and a protected human choice, so "the way platforms use this sort of technology might have constitutional significance." She separately raises whether foreign ownership and control over moderation decisions bears on whether First Amendment scrutiny is triggered at all.

Jackson, concurring in part and in the judgment. Jackson agreed on vacatur but would not have reached the merits discussion, invoking Ashwander restraint. She states two boundaries — not every action of a social-media company is protected expression, and not every hypothesized regulation would survive — and instructs courts to "carefully parse not only what entities are regulated, but how the regulated activities actually function," rejecting both category-level reasoning about social-media companies and analogy-level reasoning from prior media cases.

Thomas, concurring in the judgment. Thomas called the Court's merits discussion "unnecessary to its holding" and argued it repeats the error attributed to the circuits by selecting two functions out of "the sea of 'variegated and complex' functions that platforms perform." He argued more broadly that Article III leaves federal courts without "authority to deem a statute 'facially' unconstitutional," and said the Court should reconsider Zauderer. He joined Alito's opinion in full.

Alito, concurring in the judgment, joined by Thomas and Gorsuch. Alito agreed with the disposition and treated everything else as "nonbinding dicta." His objection is to the analogy the Court's reasoning rests on: the majority "unreflectively assumes the truth of NetChoice's unsupported assertion that social-media platforms — which use secret algorithms to review and moderate an almost unimaginable quantity of data today — are just as expressive as the newspaper editors who marked up typescripts in blue pencil 50 years ago." He allowed that the as-applied questions "are important, and we may have to decide them before too long."

Read together, Barrett's concurrence and the Alito opinion identify the same unresolved question from opposite directions: whether protection extends to curation performed by a system rather than chosen by a person. Barrett treats it as turning on how the technology is used; Alito treats the majority as having assumed the answer.

Application to AI regulation

Three strands of AI policy litigation draw on the opinion.

Output regulation. Statutes that require a model developer to carry, suppress, or label particular generated content are argued against on the editorial-control ground the opinion articulates. The reasoning is invoked in the challenges at xAI LLC v. Weiser (challenging the Colorado AI Act) and Minnesota HF 1606 (Nudification Technology Prohibition).

Disclosure mandates. The Zauderer framing governs whether an AI-labelling or explanation requirement is a permissible disclosure or an undue burden, a question that recurs across State Deepfake Statutes (MN, WA, TX, CA) and the transparency provisions of state frontier-model statutes.

Facial-challenge method. The instruction to canvass a statute's full range of applications before invalidating it facially raises the burden on pre-enforcement challenges to broadly drafted AI statutes, and is the procedural posture in which several state-law challenges now proceed. The related design-code litigation at NetChoice v. Bonta (CAADCA litigation) applies the same method in the Ninth Circuit; note that NetChoice v. Bonta is a distinct case concerning California's Age-Appropriate Design Code Act.

Commentary on the decision's carryover to generative systems is surveyed at Section 230 (47 U.S.C. § 230) and Algorithmic Speech Doctrine; positions attributed to individual scholars appear at Eric Goldman.

Relationships

• depends-on: Algorithmic Speech Doctrine
• related: Section 230 (47 U.S.C. § 230)
• related: NetChoice v. Bonta (CAADCA litigation)
• related: xAI LLC v. Weiser (challenging the Colorado AI Act)
• related: Minnesota HF 1606 (Nudification Technology Prohibition)
• related: State Deepfake Statutes (MN, WA, TX, CA)
• related: NetChoice
• related: Eric Goldman

Open questions

• Whether model outputs generated in response to a user prompt are treated as the developer's own expression, as curated third-party expression, or as neither, is not addressed by the opinion and has not been resolved in a reported decision.
• How the facial-challenge standard applies to statutes reaching both consumer chat products and enterprise APIs, which differ substantially in their expressive character, remains open.