Working on Safety at OpenAI

From Software Engineer to AI Engineer – with Janvi Kalra

Topic 8 of 10

Published May 28, 2025 · Clip 48:59–55:35 (6:36)

Jambi describes her safety engineering work at OpenAI: low-latency classifiers that detect harmful model or user behavior, measuring unknown harms as models get more capable, and integrating safety mitigations into product launches. She explains what makes OpenAI unique — the combination of startup speed and massive scale (60k requests per second on her service), an open learning culture, and passionate people. She illustrates the trust in engineers with minimal review requirements for deployments and notes how AI blurs role boundaries between PM, data science, and engineering.