The 2026 midterms are six months away, and AI-generated political deepfakes are already proliferating at a scale that 2022 would not have recognized. [1] The technology has matured. The moderation infrastructure has not kept pace.

X is the primary distribution channel. After Elon Musk's 2022 acquisition, the platform reduced its Trust and Safety staff by approximately 80 percent. [1] The synthetic media policy that existed under Twitter — which required disclosure labels on manipulated media — was not enforced consistently even before the staff reduction. It is enforced inconsistently now. The result: a platform where AI-generated audio clips of politicians saying things they did not say, and video composites that survive 24 hours before being flagged, reach millions of users before any intervention.

The satire-versus-disinformation distinction has effectively collapsed. [1] A deepfake labeled "satire" in small text functions identically to one labeled "authentic" for the 60 percent of viewers who share without reading to the bottom. And the crypto industry, which spent approximately $200 million on the 2024 cycle, is projecting similar or larger outlays for 2026, primarily through super PACs that are not required to disclose AI-generated content in political advertising. [2]

The philosophical problem Anna Arendt identified about manufactured consent applies with particular force here: when the medium for fabrication is indistinguishable from the medium for truth, the democratic precondition — an informed citizenry operating on shared facts — corrodes from below. [1] You do not need to convince anyone of a lie. You only need them to doubt everything.

No federal legislation requiring AI disclosure in political advertising has passed. The Federal Election Commission has not issued binding rules.

-- ANNA WEBER, Berlin