MIT Technology Review reports on "reverse game theory," an approach that aims to control powerful AI models by designing systems that reward them for following the rules.
Rather than trying to look inside AI models to confirm good behavior, a growing group of researchers wants incentives that make rule-following the most rational choice. One proposed trick rewards a model not just for test performance but for how well its answers predict what a second model would say, which makes deliberately underperforming a losing strategy. Princeton's Elad Hazan imagines models taking turns as solvers and auditors, though the risk of agents colluding remains unsolved.
Source: technologyreview.com, reported 11-03. https://www.technologyreview.com/2026/11/03/1147934/reverse-game-theory-mechanism-design-ai-alignment/