A federal judge has put one of California’s newest AI laws on hold, setting up a sharp test of how far the state can go in policing election deepfakes without crossing into protected political speech.
The law, AB 2839, was signed by Governor Gavin Newsom less than two weeks before the ruling. It was quickly pulled into a public fight over an AI deepfake of Vice President Kamala Harris that Elon Musk had reposted on X.
What AB 2839 Was Designed To Do
AB 2839 targets people who distribute AI deepfakes on social media when the content resembles a political candidate and the poster knows it is fake in a way that may confuse voters.
The law is notable because it does not focus on the platforms where AI deepfakes appear. Instead, it goes after the users who spread them.
Under AB 2839, California judges can order posters of AI deepfakes to remove the content. If they do not comply, they may face monetary penalties.
That structure made the law a direct attempt to deal with the spread of manipulated political media before an election. It also made the law vulnerable to a First Amendment challenge, because it placed legal pressure on individual speakers rather than only on companies hosting their posts.
The Kamala Harris Deepfake Case
The legal challenge came from Christopher Kohls, the X user who originally posted the AI deepfake of Kamala Harris. He filed a lawsuit one day after AB 2839 was signed.
Kohls’ lawyer argued in a complaint that the deepfake was satire and should be protected by the First Amendment. That argument became central to the dispute: whether California was regulating deceptive election content, or whether it was sweeping protected parody and political commentary into a broad enforcement system.
Newsom had suggested shortly after signing AB 2839 that the law could be used to force Elon Musk to take down the AI deepfake he had reposted. That comment helped turn the law into a visible dispute over online speech, political satire, and the use of AI tools in campaigns and commentary.
Why Judge John Mendez Blocked Enforcement
On Wednesday, United States district judge John Mendez sided with Kohls and issued a preliminary injunction. The order temporarily blocks California’s attorney general from enforcing AB 2839 against Kohls or anyone else, with an exception for audio messages that fall under the law.
Mendez’s reasoning focused on breadth. In his view, the law was written in a way that could reach large amounts of political expression, including satire, parody, critique, and other speech that receives First Amendment protection.
The decision recognized that California has a strong interest in election integrity and in responding to artificially manipulated content. But Mendez found that this interest did not outweigh the First Amendment problems created by the law as written.
The judge also pointed to the risk that subjective judgments about digitally altered content could chill speech. If enforcement depends on whether a post may be considered harmful or misleading, the decision suggests, state power can expand into uncertain territory around political commentary.
In plain terms, the ruling does not say AI deepfakes are harmless. It says California’s current approach, at least for now, is too broad to enforce against political speech on social media.
What The Injunction Means Now
Because the order is a preliminary injunction, it is not the final word on AB 2839. The law has been blocked temporarily, and the broader constitutional fight can continue.
Even so, the timing matters. The source article notes that the ruling makes it unlikely AB 2839 will have much effect on next month’s election.
The decision also lands in the middle of a wider push by California on artificial intelligence. AB 2839 is one of 18 new laws relating to AI that Newsom has signed in the last month.
For supporters of stronger rules around AI deepfakes, the ruling shows how hard it can be to write election safeguards that survive constitutional review. A law can target manipulated media and voter confusion, but if its terms also capture protected political speech, courts may step in.
For free speech advocates on X and elsewhere, the ruling is a major early win. After Newsom signed AB 2839, Musk and his usual allies posted a series of AI deepfakes that tested the new law.
The Larger AI Speech Problem
The case highlights a core problem for lawmakers: AI tools make it easier to create convincing political media, but political speech also includes exaggeration, parody, satire, and harsh criticism.
AB 2839 tried to draw a line around content that resembles a candidate, is known to be fake, and may confuse voters. Judge Mendez found that line too unstable for First Amendment purposes.
The result is a narrow but important pause. California cannot enforce this part of its AI election deepfake law for now, except as the injunction allows for certain audio messages. The fight over how to regulate AI-generated political content is not over, but this ruling makes clear that courts will closely examine any law that gives government officials power over online political expression.