2025: Extremist Deepfakes Spur Pivot to AI Security and New U.S. Oversight Bill

Technologies that lower the cost of persuasion or coercion—radio for fascist rallies in the 1930s, cassette tapes for Iran’s 1979 revolution, or Twitter for ISIS in 2014—have always been co-opted first by agile, under-resourced actors. Generative AI merely continues this pattern but accelerates it: a laptop now substitutes for a Hollywood studio or a professional hacking team. The simultaneous rise of on-prem solutions like EPFL’s Anyway Systems and the enterprise ‘trust shift’ recalls the post-2001 push toward on-shoring critical telecom switches after 9/11. Over a 100-year arc, the story is about the decentralization of strategic capability and the slog of governance catching up. December 2025 matters less for any single deepfake than for marking the point when policymakers, corporations and researchers publicly concede that the AI race is no longer about power but about control—a prerequisite, historically, for durable standards and, eventually, regulation that shapes the next century’s information order.