Judge Blocks California’s AI Disinformation Laws, Citing Free Speech Protections

A federal judge has struck down two California statutes designed to curb the use of artificial intelligence in political messaging, ruling that they ran afoul of the First Amendment and federal legal protections. The decision delivers a significant setback to the state’s ambition to regulate AI-driven political content in time for the 2026 election cycle.

Senior U.S. District Judge John Mendez, appointed by President George W. Bush, issued the ruling Friday. He permanently enjoined enforcement of Assembly Bill 2839, which would have prohibited AI-generated “disinformation and deepfakes” in political ads during the 120 days preceding an election. He also invalidated Assembly Bill 2655, which demanded that online platforms remove such manipulated content. In his view, both laws conflict with constitutional free speech principles and federal statutory protections, particularly Section 230 of the Communications Decency Act.

“To be sure, deepfakes and artificially manipulated media arguably pose significant risks to electoral integrity,” Mendez wrote, “but the challenges launched by digital content on a global scale cannot be quashed through censorship or legislative fiat. Just as the government may not dictate the canon of comedy, California cannot preemptively sterilize political content.”

In striking down the first bill, the judge emphasized that prohibiting AI‑generated political communication prior to an election is a preemptive restriction on speech, a dangerous path under the First Amendment’s high protection for political discourse. Though he acknowledged the genuine threat deepfake technology poses to election trust, he insisted that the government could not simply outlaw or police all manipulated content in advance. Political expression—even if misleading or satirical—often demands wide latitude to avoid chilling legitimate speech.

Regarding AB 2655, Mendez found that the requirement for platforms to remove AI-based political content conflicted with Section 230 protections. That law shields internet intermediaries from liability for third‑party content and enables platforms to moderate content without being treated as the publisher. The judge determined that California’s mandate would impose duties on platforms inconsistent with federal law. X and Rumble, both named challengers to the law, celebrated the decision as a reaffirmation of platform protections.

California Gov. Gavin Newsom had championed both measures after signing them in September 2024. At the time, he warned that generative AI posed a grave threat to democratic trust and the integrity of information shared publicly. Newsom argued the laws would guard against manipulated content aimed at distorting election outcomes.

California legislators behind AB 2839 had warned of a new breed of political deception: the ability to fabricate video or audio showing a candidate doing or saying something they never did, or to simulate robocalls or campaign mailers in the voice of a public official telling voters their polling location had changed. Their fact sheet justified the 120‑day pre‑election ban and a 60‑day post‑election restriction for content about election officials or voting systems. They also proposed requirements that AI‑altered ads be clearly labeled as manipulated and that injunctive relief be swiftly available to challenge violations.

Passage of the laws would have made California the first state to take legislated action against AI‑driven political manipulation. Yet the court’s decision now halts that experiment in the 2026 campaign environment.

The lawsuit had been brought by individuals and organizations including internet personalities and The Babylon Bee, a satirical news site. Plaintiffs argued that the laws would restrict legitimate speech—including parody, political commentary, and even critical analysis—by imposing rigid constraints before elections. They warned the broad language threatened to suppress lawful discourse.

Legal observers hailed Mendez’s opinion as a robust defense of free expression online. Though many recognize the dangers posed by disinformation and high‑fidelity AI forgeries, Mendez’s ruling underscores the constitutional limits on governmental control over political speech. The judge signaled that the remedy for false or manipulated content lies not in sweeping bans, but in enforcement after the fact—through defamation, fraud, or other targeted legal tools—not preemptive silencing.

Still, the decision may leave many Californians uneasy. The state Legislature and Newsom’s office are likely to review whether a more narrowly tailored law could withstand constitutional scrutiny. Potential revisions might focus on clearly fraudulent claims, seek to preserve parody and commentary exemptions, or target specific, verifiable harm rather than broadly banning manipulated content.

In the broader U.S. context, the ruling is likely to inform future battles over how (or whether) states can regulate AI‑enabled political messaging. The distinction between censoring before the fact and punishing false speech after the fact is now freshly drawn.

California is now blocked from implementing its AI political content rules until further action in appellate courts or new legislation. As the 2026 elections draw nearer, states and courts across the country will watch closely: can the promise of AI be balanced with the protections of free speech, or will attempts to legislate falsehood collide with constitutional walls?