Dawn Song

Dawn Song is a professor of computer science at UC Berkeley and one of the most decorated researchers working at the junction of AI and security — a MacArthur Fellow (2010), Guggenheim Fellow, ACM and IEEE Fellow, and for years the most-cited scholar in computer security. She co-directs the Berkeley Center for Responsible Decentralized Intelligence (RDI) and founded the startups Oasis Labs, built around a “responsible data economy” where people control their own data, and Virtue AI, focused on AI safety and security tooling.

Making powerful AI safe by construction

Song’s research agenda treats AI risk as an engineering problem to be solved rather than a reason to slow down. Her pioneering work on adversarial machine learning showed how models can be fooled and hardened; her recent work asks whether frontier AI helps attackers or defenders more, and pushes AI-driven formal verification — systems that ship with mathematical proofs of their own security — as a way out of the eternal cat-and-mouse game. Her Berkeley courses on agentic AI safety and security have become standard references as autonomous agents move into production.

In June 2026 Meta hired her as vice president of AI research at its Superintelligence Labs, bringing much of the Virtue AI team with her to shape safety and security for frontier models. Announcing the move, she said she was excited to advance “the safety and security of frontier AI models” — a bet that the surest path to safe AI runs through the labs building the most capable systems.