For years, we’ve been warned about the deleterious effects of deepfakes on our elections: AI-generated videos that can dupe voters into believing candidates have said or done things they haven’t. Now, well into the generative AI revolution and with the midterms months away, they’ve become a meme-y, juvenile feature of everyday politics—including in last night’s primary elections. In a tense Michigan Senate race, an attack video released by Republican candidate Mike Rogers, almost certainly made with the help of AI, depicts the Democratic nominee, Abdul El Sayed, as Odysseus. He’s shown entering Michigan in a Trojan horse alongside national Democratic icons Alexandria Ocasio-Cortez and Bernie Sanders. The video was viewed at least 2 million times on X. In the Massachusetts governor’s race, the campaign of Republican candidate Brian Shortsleeve has also released AI-generated videos maligning Governor Maura Healey. One Halloween-themed ad portrays the Democrat as a vampire in her “house of horrors,” while another imitates her voice in a mocking “radio ad.”
These are part of a broader trend. An ad released by the National Republican Senatorial Committee targeting James Talarico, the Democratic candidate for a U.S. Senate seat in Texas, features an AI-generated version of the candidate reading his past (and perhaps most polarizing) tweets aloud. Months ago, after Zohran Mamdani won the race to become mayor of New York, anonymous social media accounts used AI to generate synthetic local voters criticizing him.
That’s not to say there aren’t laws meant to curb AI’s influence. The National Conference of State Legislatures counts dozens of state statutes attempting to limit or ban the practice. The laws cover a range of scenarios, including deliberately deceptive AI and digital replicas created without a person’s consent. Some states, such as Arizona, require disclosure of AI used in political ads during election season. Others, including Minnesota, prohibit nonconsensual deepfakes intended to harm candidates. There is no federal law specifically governing deepfakes in elections.
Still, it’s hard to argue that these laws have had much impact. Many states have no restrictions at all, and it remains easy to generate an embarrassing AI video of a politician and post it online, whether it comes from a campaign, a PAC, or someone else. The laws may provide avenues for redress, but not the kind likely to change the outcome of an election. Once a video is made, it can be difficult to stop it from spreading or to ensure that disclaimers remain attached.
It’s also possible that the normalization of AI in political ads is not causing as much new damage as initially feared, even when the technology is used in crude and offensive ways. Political ads had limited efficacy even before generative AI, and politicians never needed large language models or image generators to lie about their opponents or misrepresent their positions. There is intriguing research suggesting AI could become a powerful tool for microtargeting and persuading individual voters on behalf of a candidate. But that does not appear to be where the technology is showing up in politics, at least for now.
Instead, politicians, whether the president, Andrew Cuomo, or an anonymous online poster, seem to be using the technology for its most inane application: partisan AI slop. The question is whether we’ve become inoculated to this new kind of high-tech low, and even less credulous than we were before.