Report Reveals AI Models' Biases Against Freedom of Expression

A new report from Meta's independent oversight board raises significant questions about the neutrality of AI models, concluding that several prominent language models are less willing to produce content that criticizes governments imposing restrictions on freedom of expression. This first-of-its-kind study assessed ten AI models from global companies, including Meta, OpenAI, Google, Anthropic, and China's DeepMind, evaluating their responses to requests for criticism of governments in ten different countries. The researchers categorized countries based on Freedom House ratings into 'open' and 'restricted' categories.