Anca Dragan is VP of AI Safety & Behavior at Google DeepMind and an Associate Professor in Electrical Engineering and Computer Sciences at UC Berkeley. With Rohin Shah she wrote "The case for reasoning transparency," one of the five inaugural essays of the DeepMind Institute, published September 16, 2026 (The Case for Reasoning Transparency (Shah and Dragan, DeepMind Institute, September 2026)).
Background and roles
According to her DeepMind Institute biography, Dragan leads the Google DeepMind teams responsible for content, frontier, and AGI safety and alignment, as well as for the model's behaviour and character. At Berkeley she founded the InterACT Lab, whose research the biography describes as helping to define mathematical formulations for human-AI interaction and value alignment (The Case for Reasoning Transparency (Shah and Dragan, DeepMind Institute, September 2026)). The Pacing the Frontier statement of July 2026, which she signed, lists her title as "VP, AI Safety & Alignment, Google" (Pacing the Frontier (statement from employees of frontier AI companies, July 2026)).
Research
With Dylan Hadfield-Menell, Pieter Abbeel and Stuart Russell, Dragan co-authored "Cooperative Inverse Reinforcement Learning" (NeurIPS 2016), which frames value alignment as a cooperative game between a human and a robot that does not initially know the human's reward function (Cooperative Inverse Reinforcement Learning (Hadfield-Menell, Dragan, Abbeel & Russell, 2016)), and "The Off-Switch Game" (IJCAI 2017), a formal treatment of an agent's incentive to allow or prevent its own shutdown (The Off-Switch Game (Hadfield-Menell, Dragan, Abbeel & Russell, 2016)). She was one of the doctoral advisers of Dylan Hadfield-Menell at Berkeley, with Russell and Abbeel (Source: people.csail.mit.edu).
She is one of the 41 authors of "Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safety" (Korbak et al., July 2025), which recommended that frontier developers "consider the impact of development decisions on CoT monitorability" (Source: arxiv.org).
Positions
In the September 2026 essay, Dragan and Shah argue that chain-of-thought reasoning currently gives developers "a window into the thoughts of the most powerful AI models" that should be kept open, and propose measuring chain-of-thought transparency, preserving transparent architectures through a limit on opaque serial depth, and auditing training rewards. They state that commercial goals alone "might therefore under-incentivise transparency and, in that case, regulation could help," and propose that developers deviating from standard chain of thought demonstrate comparable monitorability (The Case for Reasoning Transparency (Shah and Dragan, DeepMind Institute, September 2026)). See Chain-of-Thought Monitorability.
Dragan is among the signatories of the July 28, 2026 Pacing the Frontier statement, in which employees of frontier AI companies asked the U.S. government to support an international effort to develop the tools needed to deliberately pace the frontier of automated AI development (Pacing the Frontier (statement from employees of frontier AI companies, July 2026)).
Relationships
- related: Google DeepMind — VP of AI Safety & Behavior
- related: The Case for Reasoning Transparency (Shah and Dragan, DeepMind Institute, September 2026) — co-author
- related: DeepMind Institute — inaugural contributor
- related: Cooperative Inverse Reinforcement Learning (Hadfield-Menell, Dragan, Abbeel & Russell, 2016), The Off-Switch Game (Hadfield-Menell, Dragan, Abbeel & Russell, 2016) — co-author
- related: Pacing the Frontier (statement from employees of frontier AI companies, July 2026) — signatory
- related: Rohin Shah, Stuart Russell, Dylan Hadfield-Menell
- related: Chain-of-Thought Monitorability, Opaque Serial Depth