Chilling Political Dissent: Are LLMs Censoring Output Criticizing Restrictive Regimes?
• Meta’s Oversight Board released a 39-page report investigating "censorship-by-proxy," where foundation LLMs may be restricting content critical of speech-restrictive political regimes and their leaders. • The research examines whether AI outputs are improperly reflecting national laws that outlaw government criticism, effectively extending a regime's restrictive reach beyond its own geographic borders. • This phenomenon matters because it suggests that AI models may be enforcing illegitimate speech restrictions globally, potentially chilling political dissent and limiting information access.
aei.org