I interviewed at OpenAI for a research-adjacent role earlier this year. The behavioral portion was more substantial than I expected, and the questions had a real character to them. Sharing what I saw.
What they asked (paraphrased): Tell me about a time you disagreed with a decision and what you did about it. (Standard, but they wanted specifics. Vague 'I raised my concerns' answers didn't seem to land well based on how follow-ups went.) Describe a project where the right thing to do wasn't clear and you had to make a judgment call. What was the outcome, and would you make the same call now? Tell me about a time you had to learn something quickly because the situation required it. Have you ever worked on something where you had doubts about whether it should exist? How did you handle that?
That last question is distinctive. It's clearly connected to their stated mission around AI safety and responsible deployment. They're not looking for 'yes my conscience is pure' answers. The strongest answer I gave was honest about the tension I felt on a previous project and specific about what I did to work through it.
Values that kept coming up in subtext: Intellectual honesty. They seem to dislike candidates who oversell outcomes. Direct communication. When a follow-up interviewer pressed me on a weak point in a story, the expectation seemed to be engagement, not defensiveness. Genuine curiosity about the technology, not just the brand.
One thing that caught me off guard: They asked about my capacity to work in a high-velocity environment where direction changes often. Not in a test way, more like 'this is real, do you have a pattern for handling it?' Seemed like an honest signal about the culture.
Overall the behavioral questions felt more thoughtful than typical tech behavioral rounds. Less 'tell me about a conflict with a coworker' and more 'tell me about navigating genuine ambiguity.' Worth prepping specifically for those.