
AI safety and responsible deployment

Why PMs are suddenly valuable again | Tulsee Doshi (Head of Product, Gemini)
Topic 4 of 18
Published Nov 8, 2026 · Clip 9:58–14:10 (4:12)
Drawing on her six years leading product for Google's responsible AI team, Tulsi discusses AI safety. She praises Google's safety teams and describes safety as embedded from training through deployment. Key mitigations include deploying monitors that track the model's chain of thought to detect misalignment and enable intervention. She also highlights the under-discussed role of AI in hardening systems: Gemini 4 partners with cyber defenders, including Wiz (now a Google company), whose Scan for Good program found and helped patch a hospital system vulnerability. Asked whether AI risk is solvable, she says the path to optimism is constant, intentional, careful interrogation of how models are built.