New study reveals how AI systems enable authoritarianism in democracies

A new paper by Jeba Sania, Marta Ziosi, and Fazl Barez (arXiv:2606.17286) systematically compares six AI systems deployed in different political regimes — from the US to China — to understand how they enable authoritarianism. Drawing on academic publications, investigative reports, government procurement notices, and third-party evaluations, the researchers identify critical technical and operational features common across systems. These include centralization and co-optation of administrative data for law enforcement and political punishment, regulatory gaps that fail to deter misuse, weak user compliance that nullifies human oversight, and the encoding of protected group traits to identify vulnerable populations. Notably, these features appear in both autocratic and democratic regimes, though in varying configurations.

The study also finds that both centralized and fragmented AI system architectures contribute to authoritarianism. Centralized systems, especially those directed by security and military institutions, often bypass formal oversight. Fragmented systems, meanwhile, diffuse accountability among multiple stakeholders, allowing entrenchment of harmful practices. The authors argue that AI-enabled authoritarianism is not confined to autocracies — it is a distributed phenomenon resulting from design and operational choices made by developers, administrators, and users. The paper concludes with recommendations for policymakers and developers to mitigate these risks, emphasizing the need for stronger oversight, user compliance mechanisms, and protections for vulnerable groups.