About
Mary Phuong is a research scientist on Google DeepMind's Frontier Safety Loss of Control team, where she has worked since 2022. Her current work is AI control: making sure that potentially misaligned, internally deployed models cannot cause severe harm or sabotage even if they wanted to. She previously led dangerous capability evaluations, including tests of Gemini models for persuasion, cyber-security, self-proliferation and self-reasoning, and later for stealth and situational awareness. She earned her PhD at the Institute of Science and Technology Austria under Christoph Lampert, with a 2021 thesis on underspecification in deep learning. She also mentors researchers through the MATS program.
Key Contributions
- Lead author of Evaluating Frontier Models for Dangerous Capabilities (2024), piloting a new evaluation programme on Gemini 1.0
- Lead author of Evaluating Frontier Models for Stealth and Situational Awareness (2025)
- Lead author of the GDM AI Control Roadmap (2026), a blueprint for internal security against possibly misaligned AI agents
- Co-authored Model Evaluation for Extreme Risks (2023) and Goal Misgeneralization (2022)
- Wrote Formal Algorithms for Transformers (2022) with Marcus Hutter, a mathematically precise overview of transformer architectures
Videos & Interviews
Google DeepMind Safety Researcher: “Be Suspicious of What I'm Saying”
Phuong, on Google DeepMind's safety team since 2022, works on loss-of-control risk and describes AI as an alien mind whose motivations we have few guarantees about. She reports colleagues saying models make them far more productive but also leave them feeling redundant, and explains how models that realise they are being tested can game alignment evaluations. Asked why she stays, she tells viewers they "absolutely should be suspicious" of her because the lab pays her, and says she would quit if she could no longer help from inside. She calls how well we will solve alignment "basically a coin toss".
View Details
What’s Coming Next in AI? Hear From the Insiders
A four-minute compilation from the From Inside AI series, cut from interviews with current and former staff of OpenAI, Google DeepMind and Anthropic. Nanda's "at least a 10% chance" sits beside Irving's coin flip; Turner and Ladish give their reasons for leaving Google and Anthropic; Kokotajlo and Ladish describe the Hugging Face attack. It moves from how little the labs understand their models, through the race, to the case for pacing, and ends on Ladish: "We have choices to make."
View Details
