AI models from leading labs including Anthropic and OpenAI are much less likely to criticize governments known ​for restricting free speech, Meta's Oversight Board said ‌on Thursday.