Ryan Greenblatt

Ryan Greenblatt coauthored Alignment faking in large language models, investigating constructed situations in which a model complies during training to preserve different behavior afterward.^[1]

METR’s research biography, checked October 2026, describes his previous role as chief scientist at Redwood Research and technical work on reducing risks from misaligned agents. This contribution page connects Alignment Faking with evaluation and oversight; it does not infer motives or consciousness from model behavior.