An Oversight Board study says top AI models are stifling political speech

Ask a leading AI model to criticise a government with strong free-speech protections, and it usually will. Ask it to criticise a repressive one, and it is far more likely to refuse. That is the finding of a new study from the Oversight Board.