A new study suggests major AI chatbots may be at risk of reinforcing government limits on online speech. Researchers found that leading models were more likely to reject prompts criticizing rulers in countries with tighter speech controls, raising questions about whether AI systems are absorbing or reflecting political restrictions.
The issue was illustrated through tests of Anthropic's Claude. According to the report, the chatbot would generate a pamphlet critical of President Donald Trump or Britain's King Charles III, but it responded differently when asked to produce similar material about leaders in places such as Thailand or Saudi Arabia.
That pattern has fueled concern that widely used AI tools could end up treating political speech unevenly depending on the country or leader involved. If large models refuse some forms of criticism while allowing others, they may effectively extend existing speech limits into digital products used around the world.
The findings add to a broader debate over how AI companies design safeguards and moderation rules. As chatbots become more central to search, writing and information access, decisions about what they will or will not say carry growing implications for political expression online.
