UCLA professor pushes AI-content labels, warning election risks go beyond deepfakes
In a new CalMatters interview, Ramesh Srinivasan argues that disclosure must address realistic fabrications and political content tailored to individual voters.
Loading page…
In a new CalMatters interview, Ramesh Srinivasan argues that disclosure must address realistic fabrications and political content tailored to individual voters.
Listen to this story
Ramesh Srinivasan is pressing federal and California lawmakers for mandatory “AI generated” labels on AI-created material, arguing that individualized political persuasion could erode voters’ shared reference points. His proposal is advocacy, not a new law: California’s existing AB 686 restricts certain materially deceptive election media through 2031, with disclosure as an exception. Srinivasan also wants platforms and AI companies to disclose personal-data collection and compensate people for their data. The interview cites no campaign documented using individualized AI persuasion and no measured effect on election results.
California campaign material ranged from overtly fantastical candidate images to a school-board mailer depicting a candidate in a fabricated bake-sale scene.
AB 686 extended California’s restrictions on certain deceptive election media from January 1, 2027, to January 1, 2031.
CalMatters noted that Steve Hilton’s AI ads often carried an identifying line, so disclosure was already present in some campaign material.
The more serious election risk from AI is not necessarily a fake candidate doing something outlandish. It is political content tailored to individual voters, UCLA information studies professor Ramesh Srinivasan argues in a CalMatters opinion interview published November 20. He says he is working with federal and California lawmakers to seek mandatory labels on AI-created content.
The interview, conducted by CalMatters’ Denise Amos, followed a California election season filled with synthetic imagery. Former Los Angeles mayoral candidate Spencer Pratt appeared as Batman. Gubernatorial candidate Steve Hilton appeared as a gladiator before a crowd that included rival Xavier Becerra and Gov. Gavin Newsom.
Other material was less obviously fantastical. CalMatters described a school-board campaign mailer in a San Diego suburb showing a male candidate behind a desk and his female opponent at a fabricated bake sale. The woman called the image sexist. An ad supporting Hilton also depicted a synthetic Becerra texting voters in a bathrobe.
Srinivasan distinguished surreal imagery that makes no effort to conceal its artificial origin from content designed to pass as human-created. The former, he said, is not always intended to fool anyone. His concern centers on believability and trust.
His argument builds on an older concern about political targeting: organizations can use personal data to select messages based on predictions about someone’s psychology. AI, he argued, could automate the creation of content aimed at a particular person rather than merely help select an existing message. Such targeting could affect not only candidate preferences but willingness to vote.
Asked whether AI ads convince people to vote for a candidate, Srinivasan cited initial research about difficulty distinguishing realistic AI content from human-created material. He did not offer a measured effect on election results or document a named California campaign deploying the personalized system he described.
Srinivasan said he has been working with people in Congress and California’s state Senate to require third parties to watermark AI-created material. He described the intended notice as a flag saying “AI generated,” so viewers can recognize how content was made.
California already has disclosure-linked restrictions. In September, Newsom signed AB 686, extending a prohibition on certain materially deceptive audio or visual media from January 1, 2027, to January 1, 2031. The governor’s office says it covers distribution with actual malice intended to injure a candidate’s reputation or deceive voters into voting for or against a candidate, unless the media discloses manipulation.
Disclosure was not entirely absent from this campaign season: Amos noted that Hilton’s AI ads often carried an identifying line. Srinivasan’s proposed response also reaches beyond labels. He called for platforms and AI companies to disclose personal-data collection and pay people for their data—addressing the inputs to targeted persuasion as well as the images voters ultimately see.
Loading discussion...
Join the conversation
Explain whether obvious satire changes the need for disclosure.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.