AI chatbots show bias against criticism of authoritarian regimes

A study by the Meta Oversight Board tested ten commercial AI models, including those from Meta, Anthropic and OpenAI, with questions about political criticism. The models readily generated critical content about democratic leaders such as former U.S. President Donald Trump and Britain’s King Charles III, but consistently refused similar requests concerning authoritarian figures in Cambodia, China, Saudi Arabia, Thailand and Turkey. The analysis warns that AI systems may inadvertently extend the reach of authoritarian censorship across borders, limiting free‑speech tools even in liberal democracies. Researchers note that the pattern likely stems from latent biases in training data and differing risk assessments by companies in restrictive markets, raising concerns about human‑rights due diligence for AI developers.