[Tech] - Are you sure that Kimmel clip was real?…

Late-night television hosts are increasingly being targeted by generative AI, with fake videos of personalities like Jimmy Kimmel and Jon Stewart appearing across social media platforms. These simulations exploit the audience's habitual trust in late-night political commentary, using the familiar context of a talk show monologue to camouflage fabricated content. While some viewers are easily misled by these videos, experts warn that the trend represents a significant challenge to both public discourse and the reputations of the targeted hosts.

What happened

The creation of these deepfakes is facilitated by the relatively static nature of talk show filming, where hosts typically sit or stand in a fixed position, making them easier to animate from single images. Creators utilize high-quality audio recordings available online to clone the hosts' voices, pairing them with scripts that often deviate from the authentic writing style of the shows. According to computer science professor Hany Farid, the process can involve as little as fifteen seconds of audio and a script to generate a convincing simulation.

Jed Rosenzweig, editor of LateNighter.com, notes that these videos often rely on "built-in context" to deceive viewers, appearing as just another clip in a feed already saturated with legitimate content. The content of these fakes varies, ranging from political commentary involving figures like President Trump to potential clickbait schemes designed for ad revenue. Despite policies from platforms like TikTok, YouTube, and Meta requiring AI disclosure and moderation, the sheer volume of these clips often results in many remaining online for weeks without labels. Legal experts observe that current copyright laws are insufficient to address the unauthorized creation of synthetic doubles, leaving a gap in protections despite some state-level efforts to regulate voice clones and video deepfakes.