“A cross-lab research collaboration on "model scheming," which investigates whether AI models develop hidden objectives, relies on chain-of-thought monitoring to inspect model motivations.”