YouTube has terminated 20 channels behind a coordinated network that paid gig-economy actors to read AI-generated political attack scripts on camera.
The channels paid performers about $26 a video to deliver AI-written scripts attacking Democratic politicians, then published the clips as ordinary commentary. The network racked up 45 million views before YouTube acted. The trick was simple: swapping in a real human face and voice for an AI pipeline defeats the synthetic-media detectors most platforms rely on. YouTube only pulled the channels after journalists at Semafor, working with the detection firm Riddance AI, traced the network and published their findings.
This is the real lesson here: platform moderation tools are built to catch AI fingerprints, not AI-written scripts performed by real humans. A network can automate its writing at scale and still look authentically human on screen, which means synthetic-media detection may be structurally behind the problem it claims to solve. Election-adjacent disinformation no longer needs a deepfake to move at scale. It just needs a script generator and a gig worker willing to read it for $26.
Twenty channels down is a rounding error against 45 million views already served, and against however many similar networks haven't been caught yet.