AI Risk Values Round: Delaying a Risky Breakthrough, Worthless Equity, Persuasion

This is the culture round of an onsite loop at an AI lab that puts AI safety at the center of its hiring. Candidates describe safety as a make-or-break theme throughout the process, and some who felt their technical rounds went well believe this is the round where they failed. The interviewer mixes values questions about AI risk with behavioral questions about how you work with people. ### Clarifying Questions - For the hypothetical questions, does the interviewer want your personal position, or how you would reason as an employee inside the company's decision process? - For the behavioral questions, should the example come from your most recent role, and how much detail is expected? - How long should each answer be? ### Part 1 — Delaying a risky breakthrough Suppose your team achieved a capability breakthrough that also carried serious risk if released. Would you support delaying it? Then the follow-up: if other labs would not delay a similar breakthrough, would you still support slowing down AI progress? ```hint Make the trade-off explicit Name what is gained and lost by delaying, what would have to be true for the delay to be worth it, and what could change your answer, rather than stating a slogan. ``` #### What This Part Should Cover - A clear position, with the conditions and evidence that would change it - Engagement with the competitive-pressure argument rather than avoiding it - Concrete mechanisms (evaluations, staged release, safeguards) rather than abstractions - Consistency between the answer and how you would behave in practice ### Part 2 — If your equity became worthless If the company's stock went to zero, what would you do? ```hint Motivation under stress The question tests why you are there when the financial upside disappears; answer that directly and honestly. ``` #### What This Part Should Cover - An honest account of what motivates you beyond compensation - How the mission and the work itself factor into staying or leaving - Realism rather than performed selflessness ### Part 3 — A time you were persuaded Tell me about a time someone changed your mind on something you felt strongly about. ```hint Show the update, not just the outcome Make it clear what you believed at the start, what evidence or argument actually moved you, and what you did differently afterward. ``` #### What This Part Should Cover - A genuine, specific initial position held for good reasons - The argument or evidence that changed it, and how you evaluated it - A visible change in behavior or decision as a result - What the experience taught you about how you form and hold views ### Part 4 — Repairing a relationship Tell me about a working relationship that went badly and how you repaired it. ```hint Own your part Describe what you personally contributed to the breakdown and the specific steps you took, not only what the other person did wrong. ``` #### What This Part Should Cover - A concrete conflict with real stakes for the work - Ownership of your own contribution to the problem - Specific actions taken to rebuild trust, and their result - What you now do differently to prevent the same breakdown ### What a Strong Answer Covers - Thoughtful, nuanced positions on AI risk that show you have thought about it before the interview - Honesty and self-awareness rather than rehearsed or flattering answers - Specific, verifiable detail in the behavioral stories - Consistency across answers: the values claimed in Parts 1 and 2 appear in the behavior described in Parts 3 and 4 ### Follow-up Questions - Who should decide whether a risky capability is released: the team, company leadership, or an outside body? Why? - Describe a time you raised a concern about safety, quality or ethics that others did not want to hear. - What is one view about AI you held a year ago that you no longer hold? - If you strongly disagreed with a safety decision the company made, what would you do?