Meta’s Oversight Board Finds Top AI Models Are Hesitant to Criticize Repressive Governments

A Meta-funded Oversight Board review found that major AI language models are more likely to refuse generating politically critical content about governments in countries with restrictive speech laws than those in freer nations. The study tested 10 models from companies including Anthropic, Google, Meta, and OpenAI, revealing a higher refusal rate for content about restrictive countries. Some models cited local laws or potential risks to individuals as reasons for refusal, highlighting how AI may