US finalizes voluntary AI safety tests after OpenAI and Anthropic breaches August 3, 2026 · The Independent Read full story at source