New research challenges the idea that artificial intelligence makes dictators all鈥憄owerful
Artificial intelligence is often portrayed as a powerful new tool for authoritarian governments鈥攐ne that enables constant surveillance, predictive repression, and near鈥憈otal social control. But new research from 快猫短视频 argues that this popular narrative overstates what AI can actually do.
In The Limits of Authoritarian AI, L. Jason Anastasopoulos, Associate Professor of Public Administration and Policy, and Jie (Jason) Lian, a Postdoctoral Research Fellow at the Harvard Kennedy School and 2025 PhD in Political Science and International Affairs, 聽show that AI systems do not eliminate uncertainty for authoritarian leaders. Instead, they force regimes into difficult tradeoffs that can generate instability, backlash, and new vulnerabilities.
鈥淧eople tend to imagine AI giving authoritarian governments this all鈥憇eeing, all鈥憄owerful ability to control society,鈥� Anastasopoulos said. 鈥淏ut in reality, AI systems always make errors, and those errors create real political problems for the regimes that rely on them.鈥�
Why AI creates dilemmas, not dominance
At the center of the paper is what Anastasopoulos and Lian call a 鈥渃alibration dilemma鈥�. Any AI system designed to identify threats鈥攚hether dissenters, protesters, or political opponents鈥攎ust be programmed to decide how suspicious is suspicious enough to trigger action.
If the system is calibrated too loosely, it flags large numbers of innocent people as threats. That leads to widespread repression, public anger, protests, and political backlash. If it is calibrated too narrowly, genuine opponents slip through the cracks, organize, and potentially challenge the regime.
鈥淭here鈥檚 no perfect setting,鈥� Anastasopoulos explained. 鈥淎uthoritarian leaders have to choose between two kinds of failure. Either way, they expose themselves to risk.鈥�
The authors refer to the resulting cycles of tightening and loosening surveillance as 鈥�threshold whiplash.鈥� Governments crack down aggressively, face backlash, retreat, then tighten controls again when new threats emerge.
China鈥檚 experience during its zero鈥慍OVID policies illustrates this dynamic. Broad, AI鈥慹nabled monitoring systems flagged enormous portions of the population as potential risks, generating widespread unrest and political instability rather than seamless control.
The myth of the all鈥憇eeing state
The research also challenges what the authors describe as the 鈥�panopticon bluff.鈥� Authoritarian governments benefit when citizens believe AI surveillance is omnipresent and infallible鈥攅ven when it is not.
鈥淭he real power of AI for authoritarian leaders often lies in the perception that the system sees everything,鈥� he said. 鈥淏ut that perception is a bluff.鈥�
AI systems can be evaded through simple behavioral changes, coded language, or shifts in communication. Once people recognize the limits of surveillance, fear diminishes, and with it, some of the regime鈥檚 control.
Implications for democracy advocates
Rather than painting a dystopian future in which technology guarantees authoritarian dominance, the paper offers a more nuanced鈥攁nd more hopeful鈥攁ssessment.
The authors argue that pro鈥慸emocracy actors can take advantage of the limits of AI鈥慸riven repression by better understanding how algorithmic systems actually work and where they fall short. By demystifying these technologies, supporting individuals who are falsely targeted by overbroad surveillance, and challenging exaggerated claims about AI鈥檚 omnipotence, democratic actors can reduce the power these systems are meant to project.
鈥淚f you understand how these systems operate, you can identify their vulnerabilities,鈥� Anastasopoulos said. 鈥淎I doesn鈥檛 make authoritarian regimes unstoppable鈥攊t gives them new problems to manage.鈥�
From research to policy conversation
The article has attracted attention beyond academia. Anastasopoulos will present the paper鈥檚 arguments to staff and leadership at the National Endowment for Democracy, an organization that supports democratic movements worldwide.
The invitation reflects a broader shift among policymakers and practitioners toward recognizing that emerging technologies not only reshape authoritarian power but also create meaningful openings for democratic resistance and accountability.
鈥淭here鈥檚 a tendency to think that AI automatically means more control and less hope,鈥� Anastasopoulos said. 鈥淥ur goal is to show that the reality is more complicated, and that understanding those complications matters for people working to defend democracy.鈥�
