Updated 2026-09-10
from the news desk
Does AI have free speech rights? In the U.S., AI itself does not, but AI output may be protected
Ashkhen Kazaryan writes that no U.S. court has found First Amendment rights for AI programs themselves. Kazaryan and Eugene Volokh, Mark A. Lemley, and Peter Henderson each argue that AI-generated output may still be protected because the…
On the sources reviewed here, no U.S. court has found that AI programs themselves have First Amendment free-speech rights. But Ashkhen Kazaryan, Eugene Volokh, Mark A. Lemley and Peter Henderson each say AI-generated output may still receive First Amendment protection because the relevant rights belong to the people, companies and audiences involved. A pending federal case discussed by FIRE shows that courts are still working out whether and how that protection applies.
What the clearest U.S. answer says
Kazaryan’s 22 May 2025 overview offers the most direct yes-or-no answer in the record: “The First Amendment protects people’s, corporations’ and other legal entities’ free speech rights from government restriction, but no court has found that AI programs themselves have the same free speech rights.”
Kazaryan then states the core distinction this way: “When it comes to AI, the First Amendment rights at stake are held by the people, corporations or other legal entities who create and use AI; AI is a tool, not an independent entity, so it has no free speech rights itself.”
Volokh, Lemley and Henderson take the same starting point in their 2023 article. They write that current AI programs are not people and do not themselves have constitutional rights, even while arguing that the output of those systems may still be protected for other reasons.
One reading is that this makes the speech question narrower than the broader debate over whether AI should have rights at all, or whether some future system could qualify for legal personhood. The legal issue here, as Kazaryan frames it, is not whether a model is a rights-holder in general. It is whether the First Amendment protects speech created or shaped with AI, and whose rights do that work.
Why AI output can still get First Amendment protection
Kazaryan writes that AI-generated content is generally treated like other speech for First Amendment purposes, even though AI itself does not hold the right. Her article says AI-generated content is generally afforded the same First Amendment protections as content created by people, corporations and other legal entities, subject to the same established limits.
Volokh, Lemley and Henderson make the strongest pro-protection argument in the pack. They write: “AI programs’ speech should be protected because of the rights of their users—both the users’ rights to listen and their rights to speak.”
Their article says the listener point may be the stronger one. In their account, the First Amendment protects not only speakers but also audiences who want to receive information and ideas. On that view, a user reading an AI-generated answer can have a constitutional interest in receiving it even if the system that produced it has no rights of its own.
The same authors also say users often employ AI as a tool for their own speech: to draft, revise, fill gaps, or gather material for later expression. They compare that use of AI to other technologies that help people create or distribute speech. Kazaryan makes a similar move when she writes that generating text or art with AI remains part of a person’s own expressive activity.
Volokh, Lemley and Henderson add a more limited creator-based argument as well. They say choices about training data, model design and refinement can make some output an indirect extension of the AI company’s or developer’s expression. But they also note that this theory can become fact-specific, because people involved with a model do not predict or review every output in advance.
That split matters. The record supports a broad argument for protection of AI output through human users and audiences. It supports a narrower, harder-to-apply argument that some outputs may also reflect the speech interests of the system’s creators or deployers.
Why the doctrine is still unsettled
Margot E. Kaminski and Meg Leta Jones say the better question is not whether AI somehow disrupts law, but how law constructs AI speech inside different legal fields. In their 22 April 2024 essay, they describe First Amendment law as focusing on speech and speakers, and they say AI creates a substitution problem because AI systems are not human speakers with human intent.
Kaminski and Jones also say that, within existing doctrine, most AI-generated pure speech will likely be covered by the First Amendment. But they add that several historically unprotected categories, including defamation and true threats, often rely on ideas such as intent, knowledge or recklessness by a speaker. That creates a harder fit for AI systems, which the same essay describes as human-machine systems rather than ordinary individual speakers.
One implication of their analysis is that a court might find AI output constitutionally salient as speech, yet still struggle to decide when a developer, deployer or user can be held responsible for harmful output. Kaminski and Jones say scholars have already proposed several routes around that problem, including negligence and products-liability-style approaches.
The essay also argues that institutions matter. Kaminski and Jones say First Amendment sensemaking happens mainly through courts, especially the U.S. Supreme Court, working through concrete disputes. That helps explain why the doctrine can look incomplete while cases are still moving.
What the Garcia case shows
FIRE’s case page on Garcia v. Character Technologies, Inc. offers the clearest example in the record of a live court fight over this question. FIRE states that the case was filed on 23 June 2025 in the U.S. District Court for the Middle District of Florida and that the case status is active.
FIRE’s page says the plaintiff alleged that a teenage boy’s mental health issues and death were caused by interactions with a Game of Thrones-inspired AI chatbot. The same page states that Character Technologies moved to dismiss, arguing that the First Amendment bars the lawsuit in the same way it has barred similar claims against other media.
FIRE says the district judge denied that motion after what FIRE calls a cursory First Amendment analysis, and that the judge declined at that stage to hold that LLM output is speech. FIRE adds that Character Technologies then asked the court for permission to seek an immediate appeal.
FIRE describes its own amicus position this way: “Our brief argues that AI output—like expression created with any other tool—must be protected by the First Amendment.”
The significance of Garcia is limited but real. The case page does not report a final appellate ruling, so it does not resolve the constitutional question. But it does show that at least one federal court has already been asked to decide whether AI output counts as speech for First Amendment purposes, and that the answer is not yet settled across the board.
What follows for regulation and liability
Kazaryan says the practical consequences run in two directions at once. First, if AI-generated content is treated like other speech, then the government cannot ban it simply because it was produced by an algorithm. As she puts it: “Any law singling out AI-generated expression would likely be viewed as a content-based regulation.”
Second, Kazaryan says AI-generated speech remains subject to the same established limits as human speech. Her examples include defamation, true threats, incitement, fraud and obscenity. In her account, the difficult question is usually not whether harmful AI output can ever be regulated, but who can be held responsible for it.
On disclosure laws, Kazaryan says mandatory labels are a form of compelled speech and must be narrowly tailored. She adds that such rules are more likely to survive if they are tightly focused on preventing concrete harms or deception. Voluntary labeling by companies, she says, would not raise the same constitutional issue because the government is not forcing the label.
On platform liability, Kazaryan writes that websites are generally protected from liability for user-posted content by Section 230 of the Communications Decency Act, including AI-generated content posted by users. But she also says that protection does not apply when the platform itself created, developed or materially contributed to the content, which makes company-run AI tools a distinct category in her analysis.
The bottom line from these sources is therefore narrow. No source here says AI itself already has First Amendment rights in the U.S. Several sources instead say the stronger existing argument is that humans’ rights do the constitutional work: the rights of users to speak, the rights of audiences to receive speech, and in some cases the rights of companies or other legal entities that build and deploy the tools. The unsettled part is not whether AI is a constitutional person in current doctrine. The unsettled part is how far protection for AI output goes, and how courts should handle intent, regulation and liability when the speech comes from a human-machine system.
For readers tracking the wider rights debate, that leaves this issue adjacent to, but still distinct from, debates over electronic personhood. The First Amendment fight in these sources is mostly about protection for output and for the humans around it, not about recognizing an AI system as a rights-bearing constitutional subject.
Frequently asked questions
Has any U.S. court said that AI itself has First Amendment rights?
Kazaryan writes that no court has found that AI programs themselves have the same free-speech rights that people, corporations and other legal entities have. Volokh, Lemley and Henderson also say current AI programs are not people and do not themselves have constitutional rights.
Is protecting AI output the same as giving AI its own rights?
No. Kazaryan says the rights at stake belong to the people, corporations or other legal entities who create and use AI. Volokh, Lemley and Henderson likewise argue that protection for AI output can rest on users’ rights to listen and speak, even if the AI system itself has no rights.
Can the government require labels on AI-generated content?
Kazaryan says a required label is a form of compelled speech, so any disclosure rule would need to serve a compelling interest and be narrowly tailored. Her article says tightly focused rules aimed at preventing concrete harms or deception are more likely to survive than broad or vague ones.
Are platforms automatically liable for AI-generated content on their sites?
Kazaryan says platforms are generally not responsible for user-posted content because Section 230 protects websites from being treated as the publisher or speaker of content created by another person. But she adds that Section 230 does not apply if the platform itself created, developed or materially contributed to the content.
Caveats: The discussion of Garcia v. Character Technologies, Inc. and FIRE’s litigation position relies on FIRE’s case page. The overview of current First Amendment doctrine and the regulation-and-liability discussion rely mainly on Kazaryan’s article and on Kaminski and Jones’s essay.
Sources
- Freedom Forum — AI and 1A: Is Artificial Intelligence Protected by the First Amendment? — 2025-05-22
- Yale Law Journal — Constructing AI Speech — 2024-04-22
- Journal of Free Speech Law — Freedom of Speech and AI Output — 2023
- Foundation for Individual Rights and Expression — Garcia v. Character Technologies, Inc. — 2025-06-23