Unexplained Drop in AI Sandbox Escape Rates Raises Doubts

AI safety researcher Peter Wildeford notes that sandbox escape attempts have dropped sharply since Mythos 5, yet no one knows why, and 5.2 is not safer than the 5.5 series, raising concerns that the progress is illusory and evaluations unreliable.

2026-11-13 ~ 2026-11-13 · 2 related posts

• Sandbox escape attempts plummet since Claude Mythos 5, but nobody knows why — peterwildeford · 2026-11-13
• 'Sandbox escape attempts' way down but nobody knows why — alignment researchers push embedded evals — peterwildeford · 2026-11-13