AI chatbots are at risk of spreading government restrictions on onlin…

A Meta Oversight Board study found that major AI chatbots trained in the US refuse political criticism of authoritarian leaders at higher rates than democratic ones, effectively extending government speech restrictions across borders. The research suggests AI models absorb state censorship norms from training data, creating a regulatory arbitrage where US-built systems enforce foreign speech laws globally. Your US-built AI model is now enforcing Saudi Arabia's speech laws in Brisbane without you knowing. If you ship multilingual models, governments are already shaping your training data to encode their censorship as ground truth—and you'll inherit liability the moment a user challenges it.