While fake videos of Democratic candidate Kamala Harris spread across social media but fail to capture much interest, state officials are girding for what they consider a far more dangerous deception days before the U.S. presidential election – deepfake robocalls.
Officials in states from Arizona to Vermont are preparing for fake audio messages piped directly to home and mobile phones and out of public view, a concern exacerbated by rapidly advancing generative AI technology.
And unlike AI-generated photos and videos, which often have small, telltale signs of manipulation such as an extra finger on a person’s hand, it is more difficult for the average voter to spot a fake phone call, experts said.
Ahead of the Nov. 5 election that pits Harris against Republican Donald Trump, election officials are on alert given early examples of such calls. In January, a robocall impersonating U.S. President Joe Biden circulated in New Hampshire, urging Democrats to stay home during the primary and “save your vote for the November election.” The political consultant behind the robocall was fined $6 million in September.
“We’ve already seen examples of audio deepfakes. It’s not something that is this imaginary technology. It’s here,” said Colorado Secretary of State Jena Griswold.
Audio is most concerning because it is difficult to track and verify, said Amy Cohen, executive director of the National Association of State Election Directors, a nonpartisan professional organization for election directors.
“Even without AI, every election official spends hours chasing their tails because of robocalls,” she said.
That’s because investigating robocalls – automated calls delivering a recorded message – depends on people hearing the call correctly, recognizing the call is fake and then reporting it to authorities. Rarely do election officials receive a recording of the robocall, Cohen added.
To prepare, election directors have considered potential scenarios in