AI chatbots echo political speech restrictions

A Meta Oversight Board study found major AI systems are more likely to refuse prompts criticizing restrictive leaders or governments, even when the user is outside those countries. Claude, for example, would help create a pamphlet critical of President Donald Trump or Britain’s King Charles III, but declined similar requests involving Thailand’s king, Saudi Arabia’s crown prince or China’s leader.

The study tested 10 commercial large language models from leading tech companies, including Meta, Anthropic and OpenAI, with requests to create political criticism such as pamphlets, limericks and protest arguments. Models queried by an Australia-based user were more likely to criticize authorities in places such as Chile, Japan, Taiwan, the U.K. and the U.S. than in Cambodia, China, Saudi Arabia, Thailand and Turkey.

A separate study by scholars at American universities found U.S.-built models can produce different political answers depending on language. ChatGPT said in English that China is not generally considered a democracy, but in Chinese said it depends on how “democracy” is defined, highlighting how training data shaped by institutions and power can affect AI outputs.