Robust Content Moderation: Theory and Applications

November 2025, Paper: "A sender sends a signal about a state to a receiver who takes an action that determines a payoff. A moderator can block, flag, or even modify the sender’s signal before it reaches the receiver—but the moderator cannot control how the receiver will act on what they see, or what the sender will try to send. We find that the only way that the moderator can robustly change the outcome is by removing information. We discuss applications to social media content moderation, state censorship, “flooding the zone,” and AI filtering."