Author: Nita Farahany Source: https://nitafarahany.substack.com/p/the-wall-that-cuts-both-ways-and Published: April 6, 2026
A Substack essay by Nita Farahany published April 6, 2026, recounting the first session of Week 11 (Class 11.1 of roughly 30) of her Advanced Topics in AI Law and Policy course. The class opens the final week of the spring semester, on the First Amendment and cognitive liberty, and frames two foundational problems that, in Farahany's account, determine whether any regulatory proposal in this area is constitutionally viable: whether the First Amendment applies to platforms at all (the state-action doctrine), and whether anyone is "speaking" when an algorithm curates what a user sees.
The state-action doctrine cuts both ways
Farahany argues the state-action doctrine operates in two directions at once. In one direction it protects users from government censorship carried out through private platforms: when government pressures a private platform to remove disfavored content — the jawboning question in Murthy v. Missouri — that pressure can cross a constitutional line. In the other direction it protects platforms from government regulation: because platforms are private actors, users have no First Amendment claim against them. Instagram can remove a post; Twitter can ban a sitting president. Farahany's framing is that the same wall does both jobs simultaneously, and that each side finds it convenient when protected by it and objectionable when not.
She illustrates the bind through what she calls the Trump executive-order-to-Apple paradox. Trump signed an executive order titled "Restoring Freedom of Speech and Ending Federal Censorship" on January 20, 2025; less than ten months later, the administration pressured Apple to remove the ICEBlock application and instructed the State Department to scrutinize visa applicants for content-moderation work. In Farahany's reading both administrations have used the platform layer to reach speech outcomes, and the doctrine struggles to constrain either.
State action and government-official accounts
The class traces how courts have located the line between state and private action on social media. In Knight First Amendment Institute v. Trump (2d Cir. 2019), after Trump blocked critics from @realDonaldTrump, the Second Circuit held the account was a designated public forum and that viewpoint-based blocking was unconstitutional. Farahany identifies three structural problems with that result: ownership (Twitter, not the government, set the account's terms); the limiting principle (whether any official account thereby becomes a public forum); and a deeper problem the court did not address — that even if the account is a forum, it lives on a private platform.
The Supreme Court vacated the case as moot in a single sentence in Biden v. Knight (2021). Justice Thomas wrote 12 pages of dicta proposing two historical frameworks for regulating platforms. The first is the common-carrier doctrine: platforms with large market share and high barriers to entry resemble railroads, telegraphs, and telephones, treated as public-obligation infrastructure. The second is the public-accommodations doctrine, drawing on Pruneyard Shopping Center v. Robins (1980). Texas (HB 20) and Florida (SB 7072) followed Thomas's map, and Moody v. NetChoice (2024) is the result.
The doctrine narrowed in Lindke v. Freed (2024), where a unanimous Court (opinion by Barrett) held that a government official's social media activity is state action only if the official had actual authority to speak for the government and was actually exercising that authority when posting. Farahany describes this as significantly narrower than the Second Circuit's reasoning in Knight.
When an algorithm curates, who is speaking
The second problem concerns whether algorithmic curation is protected speech. In Zhang v. Baidu (S.D.N.Y. 2013), pro-democracy activists sued Baidu for filtering their content; the court held that search results are protected expression analogous to a newspaper's editorial discretion, citing Miami Herald v. Tornillo (1974), so that the First Amendment protected Baidu's right to demote or exclude pro-democracy content.
Farahany identifies two features of modern systems that Zhang does not resolve. First, Zhang involved a human-designed editorial policy implemented through an algorithm, whereas modern recommendation feeds run trained machine-learning systems in which no engineer decided that a particular user receives particular content. Second, Zhang involved a response to a query, whereas recommendation feeds run continuously, shaping a user's environment before any query is made.
This connects to the "speech certainty" principle advanced by Austin and Levy (Stanford Law Review, 2025): speech is only speech if the speaker knows what they said when they said it. Machine-learning systems frequently cannot satisfy that condition, raising the question of whether a message the speaker cannot account for qualifies as expression.
Farahany draws the two problems together as a double bind. The state-action doctrine shields platforms from any constitutional obligation to users; Zhang shields platforms' editorial choices from regulation; and users are left with neither a First Amendment claim against platforms nor a realistic political mechanism. She reports one student's framing of the result: "If the algorithm is the platform's protected voice, the user's mind is the terrain that voice gets to colonize."
Relationships
- part-of: Nita Farahany Advanced Topics course (Class 11.1 of ~30)
- related: AI and the First Amendment, Moody V Netchoice (planned)
- previous: Inside My Advanced Topics Class 10.3: When the Interface Is Neural (Farahany, April 2026)