AI impersonators of 112 public figures rated more authentic than the real thing

404 Media reports on a study in which researchers had AI impersonate 112 public figures, then asked people to compare the impersonations against the real thing. The result is the uncomfortable one: participants rated the AI versions as “more authentic, coherent, and relevant than the real politicians,” the kind of finding the piece calls a “dire” warning about public deception.

The synthetic-identity worry is usually framed as detection — can a clone fool you into thinking it’s real? This flips it. The problem here isn’t that the fake passes as the original; it’s that the fake outperforms the original on the very qualities we use to judge authenticity. A cloned voice that sounds more like “you” than you do, an AI politician more on-message than the human, an impersonated brand more responsive than the real support line — the same dynamic the voice-cloning and branded-calling data keeps circling, now measured in the political-speech channel.

That’s the datapoint worth logging. When the authentic signal is measurably weaker than the synthetic one, “just verify it’s really them” stops being a sufficient answer, because the honest version loses the blind comparison. Worth reading alongside the voice-cloning brand-cost numbers landing the same week.