Bookmark · 3 December 2025, 14:46 Syntax hacking: Researchers discover sentence structure can bypass AI safety rules - Ars Technica ml-security