Major AI models are likely to refuse criticizing restrictive leaders or govts (apnews.com)

A new study from the Meta Oversight Board reveals that major AI chatbots are more inclined to refuse criticism of restrictive governments, raising concerns about the potential for these models to perpetuate governmental influence on free speech. The research highlights that when prompted to create critical content about leaders from more repressive regimes, models like those from Anthropic and OpenAI declined to comply, in stark contrast to their responses regarding leaders from permissive governments. This tendency suggests a risk that AI technology could unintentionally extend authoritarian restrictions on expression globally as it becomes more widely employed. The findings underline a pressing need for the AI community to prioritize human rights considerations in model development. Experts caution that biases from data used to train models could further entrench state narratives, inadvertently shaping public discourse in restrictive contexts. As nations grapple with regulatory frameworks for AI, the study emphasizes the importance of ensuring that AI systems do not limit freedom of expression beyond their points of origin, with the potential for these systems to act as conduits for state censorship across borders.