AI's Challenges to Free Speech Doctrine

Artificial intelligence has been hailed as the most promising technology produced in decades, rivaling the Industrial Revolution. At the same time, few technologies have evoked such widespread apprehension. The rapid advance of AI has provoked the creation of a new metric—the probability of doom or P(doom)—to measure the likelihood that AI will create some cataclysm that threatens humanity.

These competing impulses increasingly inform our policy stance toward AI. On the one hand, the promise that AI will dramatically improve everything from basic productivity to education and healthcare counsels a light regulatory touch. On the other hand, fear that AI will dramatically disrupt human society argues for government intervention. Mediating this conversation is the First Amendment’s protection for the freedom of speech. AI and speech are necessarily linked. Large language models, by definition, simulate human language, learning, comprehension, and creativity.

Thus, as policymakers seek to balance the opportunities and risks posed by AI, courts are increasingly being called upon to scrutinize policy for consistency with constitutional freedom of speech principles. Courts are finding, however, that AI complicates the free speech equation, calling for cautious application of First Amendment principles.

The TikTok Dilemma

TikTok provides a case in point. TikTok is one of the world’s most popular social media applications, with over 1.9 billion users globally and 170 million in the U.S. alone. It uses sophisticated self-learning AI to provide content to users based on their individualized preferences. TikTok’s algorithm “learns” through user interactions with the platform, including videos watched, enabling it to refine and better customize its offerings to each individual. TikTok then implements that learning through interface design features, such as a simplified “For You Page” and a seamless flow of content to keep users glued to the platform.

By design, TikTok adapts to its users while voraciously capturing and storing data about them. It also operates using an algorithm controlled by ByteDance, a Chinese company beholden to the Chinese Communist Party. These features unsurprisingly raised alarms with U.S. government officials. Representative Elise Stefanik summarized those concerns well when she stated that “Communist China is using TikTok as a tool to spread dangerous propaganda that undermines American national security.”

Taking action, Congress passed, and President Biden signed into law the Protecting Americans from Foreign Controlled Adversaries Act. The Act designates as a “foreign adversary controlled application” any app operated by ByteDance and TikTok. It also bars companies from providing services allowing U.S. users to access the TikTok app unless a “qualified divestiture” occurs by which China no longer exercises control over TikTok. As a practical matter, this meant that ByteDance needed to relinquish control of the algorithm upon which TikTok depended, or else TikTok would be forced to cease operations in the United States by January 19, 2025.

Faced with this existential threat, TikTok sued and sought an injunction against the Act because it violated the Constitution. Its lead argument was that the AI-powered, ByteDance-controlled algorithm at the heart of the application is constitutionally protected speech.

Algorithms as Speech

An algorithm is not self-evidently speech. At a broad level, it comprises a set of steps, oftentimes expressed mathematically, for accomplishing a task or solving a problem. In social media, an algorithm may be described as a set of rules governing how content is filtered, ranked, selected, and recommended to users.

TikTok’s argument, however, was bolstered by the Supreme Court’s 2024 majority decision in Moody v. NetChoice. That case involved challenges brought by NetChoice, an association of web-based companies, to Florida and Texas laws that limited social media companies’ discretion to exclude content from their platforms. Lower courts reached vastly different conclusions about the constitutionality of the laws.

The Eleventh Circuit Court of Appeals found that the Florida law violated the First Amendment. The court observed that algorithmic sorting was protected speech such that “when a platform removes or deprioritizes a user or post, it makes a judgment about whether and to what extent it will publish information to its users, a judgment rooted in the platform’s own views about the sorts of content and viewpoints that are valuable and appropriate for dissemination on its site.” The Fifth Circuit Court of Appeals, in contrast, found that the Texas law met constitutional muster. The court concluded that corporations lack “a freewheeling First Amendment right to censor what people say,” and that social media platforms “exercise virtually no editorial control or judgment” over the content they post.

The courts’ differing conclusions flowed from their divergent applications of what constitutes a protected exercise of “editorial discretion” in the context of social media algorithms. Over the years, the Supreme Court has found that editors engage in protected speech when they exercise discretion about what to publish and exclude. This includes a newspaper editor who decides whether to publish an opinion article, a cable television provider who curates a lineup of channels to offer customers, and a parade organizer who picks and chooses which organizations will march in a parade. Protected editorial discretion does not, however, encompass excluding military recruiters from an on-campus job fair or solicitors from a private shopping mall.

A majority of the Supreme Court sided with the Eleventh Circuit, finding that like “editors, cable operators, and parade organizers,” the “major social-media platforms are in the business, when curating their feeds, of combining ‘multifarious voices’ to create a distinctive expressive offering.” Those offerings are “the product of a wealth of choices about whether—and, if so, how—to convey posts having a certain content or viewpoint.” The majority concluded that states have no valid interest in “changing the balance of speech” that a social media platform chooses to offer.

Several Justices, however, wrote separately to express skepticism about that conclusion. Justice Jackson observed that it was “clear” that “[n]ot every potential action taken by a social media company will qualify as expression protected under the First Amendment.” Justice Alito, joined by Justice Thomas, directly questioned whether AI-powered decision-making is entitled to First Amendment protection:

[M]any of the biggest platforms are beginning to use AI algorithms to help them moderate content. And when AI algorithms make a decision, “even the researchers and programmers creating them don’t really understand why the models they have built make the decisions they make.” Are such decisions equally expressive as the decisions made by humans? Should we at least think about this?

Justice Barrett, who joined the five-justice majority, also wrote separately to question how far First Amendment protections should extend in the age of AI-reasoning:

What if a platform's owners hand the reins to an AI tool and ask it simply to remove “hateful” content? 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”? In other words, technology may attenuate the connection between content moderation actions (e.g., removing posts) and human beings’ constitutionally protected right to “decide for themselves] the ideas and beliefs deserving of expression, consideration, and adherence.”

Barrett added: “So the way platforms use this sort of technology might have constitutional significance.”

The TikTok Decision

Arguing before the Supreme Court less than one year later, TikTok leaned heavily into the majority’s reasoning in NetChoice. At oral argument, TikTok’s counsel described its protected speech as its decision to use the ByteDance algorithm, which “displays the combination of content that we prefer our users to see on the platform.” Similarly, content creators challenging the Act argued that it burdened their right to select the “publisher” of their choice—a publisher whose algorithm, in their view, better promoted their content.

Two days before the Act’s deadline for divestment, a unanimous Court ruled that the Act passed First Amendment scrutiny. The Court acknowledged that TikTok’s “For You Page” is “shaped by content moderation and filtering decisions,” including “automated and human processes.” But the Court also found that unlike the Texas and Florida laws, the Act was not addressed to any arguably editorial activities. Rather, the Act narrowly focused on the fact that TikTok’s algorithm was owned and controlled by a foreign company beholden to an “adversary” government that retained the right to access that company’s data. As a result, the Court found it unclear “that the Act itself directly regulates protected expressive activity or conduct with an expressive component.” Moreover, the Court observed that the government supported the Act with a compelling non-speech-related justification: “preventing China from collecting vast amounts of sensitive data from 170 million U.S. TikTok users.”

The Road Ahead

Shortly after President Trump took office, he issued an executive order directing the Attorney General not to enforce the Act for 75 days while his new administration pursued a possible divestment that would allow TikTok to continue operating in the U.S. He then issued an executive order setting the “policy of the United States to sustain and enhance America’s global AI dominance in order to promote human flourishing, economic competitiveness, and national security.” To maintain that leadership role, the order calls for the U.S. to “remove barriers to AI innovation” and revokes President Biden’s mammoth (and suffocating) 2023 Executive Order calling for extensive guardrails on AI development. It also states that to maintain leadership, the U.S. “must develop AI systems that are free from ideological bias or engineered social agendas.”

As these orders were being issued, a small company in China named DeepSeek released its open-source R1 AI model. A market selloff quickly ensued as word spread that R1 was competitive with the best U.S. models despite being developed at a fraction of the cost. More troubling, users quickly observed that DeepSeek’s version of R1 aggressively censors responses to queries asking for information critical of the Chinese Communist regime.

At an AI summit in Paris a few weeks later, Vice President JD Vance urged global leaders to approach AI with a sense of “optimism,” and to avoid overregulating AI development and using it as a “tool for authoritarian censorship.” He critiqued the EU Digital Services Act for compelling companies to take down so-called “misinformation,” promised the US would block the use of AI by “authoritarian regimes” to undermine other nations’ national security through surveillance and propaganda, and cautioned against AI being “dominated by massive players looking to use the tech to censor or control users’ thoughts.”

These trends point to an immediate future in which U.S. policy will accelerate the growth and development of AI through light-touch regulation while simultaneously seeking to prevent both concentrated economic power and foreign governments from using AI to engage in censorship or propaganda. This policy stance is both optimistic about AI’s promise of revolutionary productivity growth and wary of its capacity for stifling speech and enabling mass formation. It also portends more litigation over the scope of First Amendment protection due to AI, an issue left unresolved by the NetChoice and TikTok decisions.

However, those decisions provide some nascent footholds around which to design policy. NetChoice teaches that while human-directed editorial decisions implemented through algorithms merit First Amendment protection, AI-driven editorial decisions “attenuated” from human decisions may not. TikTok teaches that content-neutral policies aimed at protecting national security may pass First Amendment scrutiny when backed by a non-speech rationale, even when those policies significantly impact AI-powered social media platforms' ability to speak at all.

These limited holdings do not constitute a unified theory of AI as speech, but they provide necessary and useful signposts for future litigation sure to come. Even more importantly, their modesty bespeaks a Court willing to move deliberately, declining invitations to issue sweeping and unnuanced rules for a technology still in its youth. For decades, AI may be the most significant driver of economic, foreign, and legal policy. Its runway is long. And for that reason, the Court should—in the words of the TikTok decision—continue to “take care” when applying old precedent to “totally new problems,” so as “not to embarrass the future.”

Ryan Bangert serves as senior vice president for strategic initiatives and special counsel to the president at Alliance Defending Freedom.