Jimmy Kimmel is currently out on summer break from his late-night talk show on ABC. But the casual fan might be forgiven for thinking the host hasn't left the building. Social media videos appear to show him commenting on everything from President Trump falling asleep in the middle of a meeting, to the president's attacks on Michigan Senate candidate Abdul El-Sayed.
Fake videos showing Kimmel and other late-night TV talk show hosts are mushrooming on social media. These videos, made using generative AI, are in some ways easier to create than deepfakes of other types of celebrity — and could potentially be more disruptive.
Why people are easily duped
Jed Rosenzweig, the founder and editor of the online publication LateNighter.com, has been investigating the rise of these deepfakes. He said late-night TV talk show hosts are especially vulnerable to having their likenesses transformed into video simulations on social media because of the ease with which audiences accept the context in which they are presented.
"Jimmy Kimmel really does go on television and talk about politics all the time, and these clips really do get posted on social media every day," Rosenzweig said. "So if you see a Kimmel thumbnail tied to something Trump did yesterday, there's a built-in context and trust there that gives [it] a kind of camouflage. It just has to look like one more Jimmy Kimmel clip in a feed that's already filled with Jimmy Kimmel clips."
Rosenzweig said there are telltale signs that these monologues are not authentic.
"The gestures are odd. The writing doesn't really sound like a live monologue. And after about 10 seconds, Kimmel disappears and the cloned voice just runs over news footage," he said.
Yet some of these videos rack up hundreds of thousands of views and elicit comments that suggest at least some people think the content is real. One viewer commented on a YouTube video: "Thank you Jimmy for speaking the truth."
Threats to free speech
Rosenzweig said while other hosts like Jon Stewart are also victims of these AI simulations, the Jimmy Kimmel Live! deepfakes are of particular concern because of the host's especially contentious relationship with the Trump administration.
President Trump has repeatedly called for Kimmel to be fired. Rosenzweig said the fake videos, many of which make fun of or criticize Trump, could strengthen his administration's resolve at a time when the Federal Communications Commission and ABC are in a legal battle over free speech rights. "ABC are being scrutinized on a nightly basis for what Jimmy Kimmel is saying," Rosenzweig said.
ABC did not respond to NPR's request for comment.
Easy to create
These AI-generated talk show hosts as a genre in some ways pose a bigger threat than other kinds of celebrity deepfakes, such as of singers and actors: They are usually easier to create because these hosts typically just stand or sit in one place and talk to the camera, and there are plentiful high-quality audio recordings of their speaking voices available online to clone.
"In many cases, I can take a single image from a late show host and just animate it," said Dartmouth College computer science professor Hany Farid, who specializes in analyzing deepfakes. "Here's a voice, fifteen seconds. Here's a script. Go."
Also, Farid said, late-night TV is where many Americans get their news and validate their opinions.
"The deepfakes we are seeing, it's essentially disinformation campaigns that now seem to be coming out of the [voices] of those very people that we're used to listening to," Farid said. "So they're quite effective."
The allure of Late Night … online
Even though the audience is in decline, roughly a quarter of Americans still watch late-night talk shows regularly. According to a 2025 survey from the Associated Press-NORC Center for Public Affairs Research, the majority of these viewers are Democrats, and more of them are watching clips online than the full shows on regular linear TV. Farid and Rosenzweig both said this may explain why it's harder to find AI deepfake monologues from right-leaning talk show hosts on social media.
"The things I've seen are more on the side of scams and fraud," Farid said. "It's like, 'Sean Hannity says you should buy this cryptocurrency!'"
Representatives for hosts including Jimmy Kimmel, Jon Stewart, Sean Hannity and Greg Gutfeld declined or did not respond to NPR's requests for comment.
The handful of creators who produce talk show host deepfakes did not respond to NPR's questions about their motives.
Farid said they could be doing it for ideological reasons — or to make money.
"It's pure clickbait. Get eyeballs. Get ads. That's it," Farid said.
Platforms respond
The platforms that host this AI slop are trying to police it. TikTok, along with Google, which owns YouTube, and Meta, which owns Instagram, all sent statements to NPR saying they require creators of deepfake content to label it as being made with AI. They said they use likeness detection and moderation technologies to find unauthorized deepfakes as well as provide ways for victims of these simulations to request takedowns themselves.
"We tackle low-quality or deceptive AI content using systems we've developed over years to fight spam and clickbait," said the statement from Google, of which YouTube is a subsidiary. "As this technology advances, we've developed industry-first tools to protect the likenesses of public figures and civic leaders, and we automatically apply labels to realistic AI content for viewer transparency."
"We require people to disclose, using our AI-disclosure tool, whenever they post organic content with photorealistic video or realistic-sounding audio that was digitally created or altered, and we may apply penalties if they fail to do so," said the statement from Instagram and Facebook parent company, Meta. (Meta required NPR to submit an example of a late-night talk show deepfake video from Instagram or Facebook in order to respond to questions. The company confirmed that it took the page, which featured Kimmel, down – after being notified by NPR.)
"TikTok's policy is to remove harmful deepfakes, and separately, to require labels on realistic AI content that is not otherwise violative," said the statement from TikTok.
But the sheer speed and volume with which these videos pop up means many slip through.
One Jon Stewart deepfake about Italian Prime Minister Giorgia Meloni has been up on TikTok for more than two months. It does not have a visible AI label. Similarly, a video on YouTube featuring a deepfake Kimmel talking about Trump boasting about his prowess on the golf course was still up online at the time of publication — three weeks after being posted. It has more than 230,000 views.
Laws lag behind
Besides the platforms' whack-a-mole attempts to get rid of the deepfakes, there are currently few legal remedies to counteract the issue.
"The issue of deepfakes for things like late-night TV hosts falls into a space that is a little bit of a gap in the law right now," said Louis Tompros, an intellectual property litigator who handles artificial intelligence cases.
Tompros says existing copyright law only combats unauthorized uses of creative works, not unauthorized synthetic doubles of creative people. However, a few states such as California and Tennessee do provide protections against AI-generated voice clones and video deepfakes. A growing number of states have enacted guardrails or now require disclosures for deepfakes used in political ads; the federal TAKE IT DOWN Act, signed into law last year, criminalizes the non-consensual publication of intimate deepfakes. But that doesn't cover deepfakes of non-sexual content, like talk show hosts' commentary.
Meanwhile, several late-night hosts, such as John Oliver, have spoken out about the dangers of deepfakes on their shows. The technological downsides also provide fodder for comedy.
