A Meta Oversight Board study found that large language models used for AI chatbots may refuse or avoid producing criticism of restrictive leaders more often than models generate political critique in permissive contexts. The report warned that without human-rights due diligence and mitigation, AI systems could extend illegitimate restrictions on freedom of expression. The study tested commercial models across political criticism scenarios and found a pattern: responses were more likely to generate political criticism where it is legally restricted and penalized when the target authorities sit in permissive environments. For higher education, the findings intensify scrutiny on how AI tools are deployed for student services, research, and learning—especially where campus policies rely on third-party AI systems and automated content filters.