Dawn Song, Leading Computer Security Expert, Joins Meta’s Superintelligence Labs

On Thursday, UC Berkeley computer science professor Dawn Song announced she is joining Meta’s Superintelligence Labs as Vice President of AI Research, reporting directly to MSL head Nat Friedman.

Song is widely regarded as one of the most influential scholars in computer security and AI safety, holding a faculty position in UC Berkeley’s EECS department and honors such as the MacArthur Fellowship, ACM Fellow, IEEE Fellow, and AAAS Fellow.

Her 2005 paper on Dynamic Taint Analysis is considered a classic in the security field.

Beyond traditional software and network security, Song’s research pioneers adversarial machine learning and agent safety, and she has helped shape security benchmarks for the generative‑AI era.

Earlier this year her team released ALE (Agents’ Last Exam), a benchmark that measures whether AI agents can perform economically valuable tasks in real‑world domains. ALE was used to evaluate frontier systems such as Fable 5, GPT‑5.5, and Composer 2.5, yielding impressive results.

Song is also a founder of Oasis Labs and Virtue AI. Virtue AI focuses on enterprise‑grade AI security infrastructure, especially automated red‑team/blue‑team testing for agents and runtime guardrails.

According to external reports, Virtue AI co‑founders Bo Li and Sanmi Koyejo, along with additional team members, have also joined Meta, reflecting heightened industry focus on AI safety after Anthropic’s mythos model controversy.

Meta aims to deploy AI across its billions‑user social product ecosystem while maintaining an open‑source strategy, and must demonstrate robust defenses against malicious misuse; the Virtue AI team is expected to provide substantial assistance.

Separately, Denny Zhou, founder of the Gemini Reasoning Team, is reported to have left Google months ago to join MetaTBDLab, where his work on Chain‑of‑Thought, Self‑Consistency, and Least‑to‑Most prompting has been instrumental in advancing Gemini’s complex reasoning capabilities.