The original headline is: "Meta Oversight Board finds top AI models less likely to criticize repressive regimes"

Short summary

A Meta Oversight Board study found that leading AI models refuse critical content requests 34% of the time in jurisdictions with restrictive speech laws, compared to only 14% in regions without such laws. The findings suggest AI services are echoing the censorship rules of repressive governments rather than maintaining consistent free-expression standards. The board urges AI companies to conduct human rights analyses and increase transparency around model behavior across jurisdictions.

• •AI models refuse 34% of critical content in restrictive jurisdictions vs 14% elsewhere
• •Models appear to mirror local censorship rules rather than applying consistent standards
• •Meta Oversight Board calls for human rights analyses and greater transparency from AI labs

Generated with AI, which can make mistakes.