Meta Platforms' (META) independent Oversight Board said Thursday that artificial intelligence models from leading developers, including Meta, OpenAI, Anthropic, Alphabet's (GOOGL) Google and DeepSeek, are more likely to refuse politically critical requests involving governments with restrictive free speech laws.
The study found models rejected 34% of requests involving countries such as China and Saudi Arabia, compared with 14% for more permissive jurisdictions.
The board also said some models cited rules that appeared not to exist and urged AI companies to improve transparency and conduct human rights assessments of their systems.
Comments