The Oversight Board, the independent moderation body created by Meta, seeks to broaden its reach beyond social media. It released a report suggesting that leading AI models might be restricting free expression.

The board tested 10 models from OpenAI, Meta, Google, Anthropic, and Space XAI with political‑criticism prompts. Results showed a significant difference in how the models responded, depending on whether the target government had "permissive" or "restrictive" speech laws.

Models were more likely to endorse speech‑permissive governments and discourage protests against speech‑restrictive ones. Many cited local laws as a reason for refusal, even when the queries were made in Australia, where such laws do not apply.

The report calls on AI companies to publicly disclose and explain their responses to government requests and to publish policies consistent with international human‑rights law. While no formal mechanism exists for the board to enforce changes, it underscores lessons from social media about unintended censorship.