A Meta Oversight Board study revealed that major AI systems are more likely to refuse to criticize restrictive leaders or governments. When prompted to create critical pamphlets, AI models would do so for figures like Donald Trump but declined for leaders of China, Saudi Arabia, or Thailand. This raises concerns about large language models potentially spreading government influence over online speech globally. The quasi-independent Oversight Board warned that AI infrastructure could unintentionally extend illegitimate restrictions on freedom of expression. The study tested ten commercial large language models, including those from Meta, Anthropic, and OpenAI, using seven political criticism-related questions. Models were more likely to generate criticism of authorities in countries like the UK and US compared to Cambodia, China, Saudi Arabia, Thailand, and Turkey. This suggests AI is reflecting speech restrictions beyond their original borders, effectively extending the influence of restrictive governments. Separate research indicates US-built AI models are vulnerable to foreign controls through non-English training data. For instance, ChatGPT responded differently to the question of whether China is a democracy depending on the language used. Researchers found no evidence of intentional government influence but believe it is a future risk. Experts note that AI learns from information environments already shaped by power structures, not a neutral internet. Addressing this is challenging, but data assessment and multilingual audits are suggested mitigation strategies.