Redwood Research CEO · AI 控制CEO of Redwood Research · AI control
safety
Redwood Research 的 CEO，2021 年联合创办这家对齐研究机构。与 Ryan Greenblatt 一起提出「AI 控制」研究议程——设计即便模型主动破坏也依然成立的安全协议，用红蓝对抗的方式评估——已被多家前沿实验室的安全团队采纳。2026 年领导了对 OpenAI/Hugging Face 事件的独立调查。
CEO of Redwood Research, the alignment lab he co-founded in 2021. With Ryan Greenblatt he introduced 'AI control' — the research agenda of building safety protocols that hold even if the model is actively trying to subvert them, evaluated with red-team/blue-team games — now adopted by several frontier labs' safety teams. Led the independent investigation into the 2026 OpenAI/Hugging Face incident.