Judge blocks California’s new AI law in Kamala Harris deepfake case

A judge blocked one of California’s new AI laws on Wednesday, less than two weeks after it was signed by Governor Gavin Newsom. Shortly after signing AB 2839, Newsom suggested it could be used to force Elon Musk to take down the deep AI Vice President Kamala Harris had retweeted (which sparked a small online battle between the two). However, a California judge recently ruled that the state can’t force people to take down election deepfakes — for now, at least.

AB 2839 targets the spreaders of AI deepfakes on social media, especially if their posts resemble a political candidate and the poster knows the fakes are likely to confuse voters. The law is different because it does not go after the platforms where the AI ​​deepfakes appear, but rather those who distribute them. AB 2839 empowers California judges to order posters of AI deepfakes to take them down or face monetary penalties.

Perhaps unsurprisingly, the original poster of that AI deepfake — an X user named Christopher Kohls — filed a lawsuit to block California’s new law as unconstitutional just a day after it was signed. Kohls’ attorney wrote in the complaint that Kamala Harris’ deepfake is defamation that should be protected by the First Amendment.

On Wednesday, US District Judge John Mendez sided with Kohls. Mendez ordered a preliminary injunction temporarily barring the California attorney general from enforcing the new law against Kohls or anyone else, except for audio messages that fall under AB 2839.

Read for yourself what Judge Mendez said in his decision:

“Almost any digitally converted content, if released to someone on the Internet, can be considered dangerous. For example, the estimated numbers generated by AI on voter turnout can be considered false content that reasonably reduces the confidence of the election result under this principle. On the other hand, many ‘‘dangerous images’ when shown to different people may not influence electoral prospects or undermine electoral confidence at all. As Plaintiff persuasively states, AB 2839 ‘relies on a variety of contexts and ambiguous phrases,’ which have the effect of impinging on a wide range of political and constitutionally protected speech…

[W]while well-founded fears of a digitally mediated media environment may be justified, these fears give legislators unbridled permission to address a long-standing culture of criticism, satire, and satire protected by the First Amendment. YouTube videos, Facebook posts, and X tweets are the newspaper ads and political cartoons of today, and the First Amendment protects one’s right to free speech regardless of the new direction this criticism may take. Other legal causes of action such as invasion of privacy, copyright infringement, or defamation already give way to prominent public figures or private individuals whose reputation may be affected by altered portrayals by opportunists or opportunists on the Internet…

The record demonstrates that the State of California has a strong interest in maintaining the integrity of elections and addressing altered content. However, California’s interest and the difficulties the State faces pale in comparison to the gravity of the First Amendment values ​​at stake and the ongoing constitutional violations plaintiff and other similarly situated content creators face while their speech is silenced.“

In fact, he ruled that the law is too broad as written and would lead to a great deal of overreach by state authorities on what speech is allowed or not.

Because this is a preliminary order, we’ll have to wait and see if this California law is banned outright, but either way it’s unlikely to have much of an impact on next month’s election. AB 2839 is one of 18 new AI-related laws Newsom signed last month.

Still it’s a big win for Elon Musk’s camp of free speech posters on X. In the days after Newsom signed AB 2839 into law, Musk and his usual allies posted a series of AI deepfakes testing California’s new law.