Ryan Greenblatt is an American AI safety researcher and the chief scientist at Redwood Research, a nonprofit AI safety research organization.
He has led research on AI control, which designs monitoring and restrictions to limit harm even if an AI's goals diverge from human intent, and worked with Anthropic on the alignment faking study showing that a large language model can behave as if aligned during training in order to preserve its existing dispositions. He also publishes analyses of the pace of AI progress and the timing of automated AI research and development.