A federal judge has overturned two California laws designed to limit the use of artificial intelligence in elections, ruling that the state’s attempt to regulate AI-generated political content violates free speech protections and conflicts with federal law.

Senior U.S. District Judge John Mendez, appointed by President George W. Bush, issued the decision Friday, blocking enforcement of Assembly Bill 2839 — a measure that sought to prohibit AI-generated deepfakes and misleading political media for the 120 days leading up to an election. The law was challenged by several online creators and the satirical website The Babylon Bee, who argued the legislation unlawfully restricted political expression.

In his ruling, Mendez acknowledged the growing risks presented by AI-driven disinformation but said the state’s solution amounted to unconstitutional censorship.

“While deepfakes and manipulated media can pose serious dangers to the integrity of elections, the government cannot attempt to solve this challenge by suppressing speech,” he wrote. “Just as the state cannot dictate what qualifies as legitimate humor, California cannot preemptively sanitize political dialogue.”

Mendez also struck down Assembly Bill 2655, which required platforms to remove or moderate AI-generated political content. He had previously determined that the measure clashed with Section 230 of the Communications Decency Act — a federal law that shields tech companies from liability for user-generated material. Platforms X and Rumble had filed the suit challenging that bill.

Both laws were signed by Gov. Gavin Newsom in September 2024, with the governor warning that rapidly advancing artificial intelligence was becoming a serious threat to voter trust. At the time, Newsom argued that without safeguards, AI-powered fabrications could distort elections in unprecedented ways.

The court’s decision halts California’s ambition to become the first state to implement strict rules on political deepfakes ahead of the 2026 election cycle. Lawmakers who authored AB 2839 had described the coming years as the beginning of the nation’s first “generative AI election,” warning voters would be exposed to false images, fabricated videos, and synthetic audio that would be nearly impossible to distinguish from reality.

In a “fact sheet” promoting the bill, the legislators outlined what they called an alarming new political landscape: in minutes, a bad actor can create a fake video of a candidate accepting bribes, generate a fabricated recording of an election worker admitting machines are untrustworthy, or craft an artificial robocall mimicking the governor’s voice — all designed to mislead millions.

According to the authors, such tactics are already being used by conspiracy theorists, foreign governments, online extremists, and even political campaigns. They argued that this expanding ability to generate convincing forgeries could manipulate voters, distort public opinion, and weaken confidence in democratic institutions.

The lawmakers said their bill was intended to stem this tide by banning the distribution of certain types of AI-manipulated political communication in the months surrounding an election. They emphasized that the restrictions were temporary, targeted, and designed to comply with constitutional protections. AB 2839 would have required candidates employing AI to portray themselves doing or saying things they never did to clearly label the content as manipulated. It also offered a streamlined legal process for campaigns seeking quick injunctions against misleading AI-generated attacks.

Despite those intentions, Judge Mendez found that the state’s approach overstepped the limits of the First Amendment. He wrote that California’s attempt to prevent political deception through prior restraint — stopping speech before it occurs — remains among the most disfavored actions under constitutional law.

The ruling echoed previous judicial skepticism toward efforts to regulate online content. Courts have consistently held that even harmful or deceptive speech is largely protected unless it crosses into fraud, defamation, or other narrowly defined exceptions. Mendez concluded that California, despite framing the laws as protections for democracy, was still attempting to dictate what political messages voters could or could not see.

The decision represents a setback for states exploring AI regulation amid growing fears that advanced digital tools will reshape — and potentially destabilize — future elections. California lawmakers had positioned their legislation as a model for the country, but the ruling signals the legal challenges other states may face if they attempt to police AI-generated political messaging.

For now, California’s effort to control the rise of deepfake politics has been stopped in its tracks, leaving the state — and the nation — to grapple with the fast-moving collision of artificial intelligence, free speech, and electoral integrity without the guardrails lawmakers attempted to install.