Jimmy Kimmel and Jon Stewart Are Being Mass-Deepfaked, and Almost Nobody's Taking Them Down

A YouTube deepfake of Kimmel discussing Trump's golf game had racked up more than 230,000 views three weeks after it was posted, and was still online at the time NPR reported on it ^[1]. The Stewart-Meloni fake had been circulating on TikTok for more than two months without any AI-disclosure label, despite platform policies that require one ^[1].

Dartmouth computer scientist Hany Farid, who studies synthetic media, told NPR the format makes late-night hosts unusually easy and dangerous to fake: "In many cases, I can take a single image from a late show host and just animate it," since hosts sit or stand still while talking and abundant high-quality audio of their voices already exists online for cloning ^[1]. Farid called the pattern "disinformation campaigns" that exploit the trust audiences place in familiar on-air voices ^[1].

The volume problem is bigger than these two hosts. Reported deepfake incidents jumped roughly 982% in 2025, and Americans lost an estimated $893 million to AI-driven fraud over the same period, according to data compiled by BrokerChooser ^[4].

ABC declined to comment when NPR asked about the deepfakes, and representatives for Kimmel, Stewart, and even frequently-deepfaked conservative hosts Sean Hannity and Greg Gutfeld either declined or didn't respond ^[1]. No one whose likeness is being used has publicly addressed it.

The timing raises the political stakes: Trump has repeatedly called for Kimmel to be fired, and ABC is currently in a legal dispute with the FCC over free-speech obligations ^[1]. Fake Kimmel clips that mock or criticize Trump could, ironically, be read by the administration as evidence of network bias — even though Kimmel had no part in making them.

Media analyst Adam Rose argues the deeper danger isn't any single fake video but the erosion of the public's ability to trust any video at all — pointing to unrelated 2026 incidents like a White House-shared, digitally altered arrest photo and a disputed AI image Trump posted of Nicolás Maduro in handcuffs as evidence the problem now cuts across the political spectrum ^[3]. Rose's proposed fix is authentication infrastructure — cryptographic "content credentials" baked into cameras at the point of capture — rather than after-the-fact fact-checking ^[3].

FALSE. Every one of the specific Kimmel and Stewart clips described here is a fabrication — the hosts did not say the things attributed to them. This isn't a case of missing context; it's synthetic audio and video generated without consent and, in the case of the Stewart-Meloni clip, distributed for months without the AI label platforms require ^[1].

The bigger takeaway for viewers: a clip of a recognizable media figure "reacting" to news, with no link to an official channel or verified account, should be treated as unverified by default — especially when it's a talking-head monologue format, which Farid notes is now trivial to fake ^[1].