In this episode of Women in Product, host Shannon Peavey sits down with Tonia Osadebe, AI Safety Lead at OpenAI, for a candid and practical conversation about what it really means to build responsibly in fast moving AI environments. Tonia shares her path into AI Safety, from her early work on AI Agents and machine learning fairness to stepping into a role focused squarely on evaluating and mitigating risk in frontier systems. She reflects on her time at Google, including being part of the team navigating the widely discussed “Glue on Pizza” AI search result moment, and what those high visibility incidents teach teams about iteration, accountability, and resilience. At the heart of this conversation is a simple but powerful idea: safety work is collaborative. Tonia explains how cross functional teams come together to define acceptable risk, make principled tradeoffs, and agree to improve systems over time rather than striving for perfection before launch.

02:29 Where does safety start in a project?

03:50 We’re mitigating - not eliminating - risk

08:14 Trying not to break everything

09:26 Let’s talk about glue on pizza

11:34 What we don’t know YET

14:06 How teams collaborate

19:02 We are not the “fun police”

21:35 Tension is expected

22:15 Measure where you can

24:05 Sharing ownership

25:15 Reflecting the real world

29:00 Access will change the way we dream

AI Safety in Practice with OpenAI's Tonia Osadebe