The Atlantic's quietly alarming piece on AI as a tool for authoritarian statecraft and a new arXiv paper arguing that AI systems used in foreign policy are dramatically under-evaluated arrived on the same day. That is not a coincidence so much as a convergence that has been coming for years. We just weren't paying attention.
The Evaluation Gap That Autocracies Will Exploit First
The paper, "The Foreign Policy AI Evaluation Gap" by Pozniak and Sania, makes a pointed argument: AI systems deployed in statecraft contexts face almost none of the benchmarking rigor applied to, say, medical AI. There is no equivalent of the FDA for geopolitical LLMs. Meanwhile, The Atlantic documents how Chinese chatbots are already trained to deflect, deny, and redirect on Tiananmen, Taiwan, and Xinjiang with a fluency that human censors could never scale to. The asymmetry is stark: democratic governments are still debating AI governance frameworks while authoritarian ones are shipping product.
When the Feed Is the Foreign Policy
This connects uncomfortably to the hacktivists who defaced U.S. Army websites this week, and to BlackRock's AI credit thesis assuming AI investment is structurally sound. The attack surface for information warfare is now every interface where a person types a question and trusts the answer. A 2023 paper in Journal of Democracy by Kendall-Taylor and Frantz found that digital tools have extended authoritarian durability by an average of several years. AI does not change that finding. It accelerates it. The chronically online already know that the feed shapes belief. The question is who is shaping the feed, and for whom. Marcus Bösch's concept of "vibocracy", where affect outpaces argument and reality dissolves into circulation, is no longer a theoretical framework. It is Chinese foreign policy.