Artificial-intelligence-generated propaganda supporting the Palestinian militant group Hamas is deliberately targeting an American audience, according to a deepfake detection company studying online influence efforts that employ cutting-edge AI technology.

Investigators at Sensity AI say they have uncovered manipulated online content backing both sides of the Israel-Hamas war, but the intended audiences for each side appeared far different, providing a rare window into whom the deepfake creators seek to engage and persuade.

“Hamas propaganda targets were definitely [Europeans] and American citizens,” Francesco Cavalli, co-founder of Sensity AI, said in an email. “Israeli propaganda tried to focus more on influencing the internal audience, not only Israeli people (in order to gather support for the military operation in Gaza) but also Palestinians who are not supporting Hamas.”

---

---

Sensity AI’s report on deepfakes said a “massive propaganda conflict” is playing out on social media between partisans of Hamas and Israel.

The Amsterdam-based company said pro-Hamas content used AI to generate images of Gaza residents, especially children, next to rubble and victims of alleged Israeli attacks. The pro-Israel side, meanwhile, depicted large crowds waving Israeli flags and used deepfake videos and voice-cloning tools to portray celebrities expressing their support for Israel.

Sensity AI said its findings heightened concerns about the AI tools affecting the U.S. elections in November, but much of what the company found in the early stages of the 2024 campaign could be categorized as concerning or satirical, depending on the audience’s perspective.

A deepfake video showed Florida Gov. Ron DeSantis dropping out of the presidential race before he officially ended his campaign for the Republican Party nomination. The video contained a watermark of a meme maker who backs former President Donald Trump and concluded with a talking cartoon meatball that made clear the manipulated video was coming from an anti-DeSantis producer.

Mr. Cavalli said only a few of the thousands of AI-generated audio, images and videos about the election his team has observed ought to be considered genuine threats. Still, the less-informed may not understand the satirical hallmarks of manipulated content, he said, and other misleading content is more sophisticated.

Sensity AI is selling its ability to detect deepfakes. According to Amazon Web Services’ marketplace, 12 months of Sensity AI’s audio, image and video detection tools cost $90,000.

Election fears

Some lawmakers on Capitol Hill say they fear their political opponents will use emerging generative AI tools to challenge them in November.

Last month, the Senate Rules Committee advanced three bills focused on AI and elections, including legislation restricting deepfakes about political candidates, mandatory labeling of AI in political ads, and a push to develop voluntary guidelines for election offices.

Sen. Amy Klobuchar, Minnesota Democrat, is leading the charge to pass restrictions about AI use in elections into law, but the likelihood of anything gaining momentum grows increasingly less likely as the elections approach.

“We are going to see this resurgence of fakery and scams going on in our elections,” Ms. Klobuchar said at a committee meeting last month. “And whether you’re a Democrat or Republican, whether you’re conservative or liberal, we cannot have our democracy undermined by ads and by videos and by robocalls when you literally don’t know if it’s the candidate you love or the candidate you dislike. We cannot have that happen to our democracy.”

Politicians telling lies on the campaign trail is not new, but measuring whether deepfakes directly affect anyone’s vote is difficult.

Some foreign propaganda efforts are not designed to change public opinion but to amplify divisive sentiments and issues to drive up anger.

The Department of Homeland Security, the FBI and the Office of the Director of National Intelligence published guidance in April urging election officials across the country to be alert to the use of AI to manipulate voters.

“Train staff on standard procedures for responding to suspected AI-generated media and understand the mechanisms for notifying members of your organization about this activity,” the agencies’ guidance said.

Other guidance encouraged voters to verify sources of information and directed audiences to government websites for accurate information.

“Foreign malign influence agents engage real people with the goal of having them echo [their] messaging, essentially co-opting their already established online megaphone, while hiding the foreign origin of the influence message,” the agencies said. “Many of these tactics are not new, but generative AI tools have made it much easier and cheaper to generate and spread convincing foreign malign influence content.”

The agencies encouraged those responsible for administering elections to consider using “safe words” for authentication to ensure they are not duped by AI-powered impersonators.