A practice prompt we wrote. No company or candidate report names it, so it carries no company tag.
How to answer
There is little public evidence of how employers test this, but two sources point at it. In a Blind thread about Anthropic Applied AI interviews, one commenter advised, in September 2026, having good stories ready about AI safety and personal values. Source 1Anthropic Applied AI interviewsPublisherBlindSource typecandidate report on Blind One prep site, interviewing.io, says Palantir looks for people comfortable discussing civil liberties and rights, and cites no data. Source 2Palantir’s Interview Process & QuestionsPublisherinterviewing.ioSource typeinterview prep site Neither describes a round or how answers are judged. The test worth preparing for is whether you can turn unease into something a customer can act on.
- The request, and why it was reasonable. Who asked, for what outcome. A plainly illegal request needed no judgment.
- The specific harm. Who could be hurt, how, and the evidence you found: a feature weight, a data field, an output. “The model ranks people higher when their notes mention a death” is a concern; “I had privacy worries” is not.
- When and to whom. Early, relative to the build. To the owner first, and your own manager or account lead the same day, since your company’s name is on what ships. Know what would send you to compliance or legal first: a legal breach, or an owner who is the source of the harm.
- Your proposal. A way to meet their goal without the harm, and its cost.
- What happened. Who decided, and whether you agreed. If overruled, what you did next.
- What you check for now.
Quote the sentence you used to raise it, so the listener can judge whether the owner would have heard a proposal or an accusation.
Two traps. The hero story, where you alone saw it, makes you sound hard to work with. And a concern raised only after launch, once the harm was public and agreeing cost nobody anything, shows no judgment under pressure.
Follow-ups
What the interviewer may ask next, once your first answer is on the table.
- What would you have done if the answer had been no?
- How did you know the concern was a real harm and not your own discomfort?
- Who else could see the problem, and why hadn’t they raised it?
- Is there something you built and went along with that you would now push back on?
- Why the head of collections before compliance? What would have made you go to compliance first?
- When did your own company hear about it, and what would you have done if they had told you to drop it to protect the account?
Where answers go wrong
- Raises the concern as a feeling (“it didn’t sit right”) with no specific harm, no evidence and no alternative, so nobody could have acted on it.
- Tells a story in which only the speaker had a conscience, or in which the concern surfaced after launch, when raising it cost nothing.
Answer this in two minutes
Write the answer you would say out loud. The clock starts with your first word.
Illustrative answer about a fictional project
I was building a call-prioritization tool for the collections team at a consumer lender. The head of collections wanted each agent’s morning list ranked by how likely a borrower was to pay if called that day, trained on two years of outcomes and the agents’ free-text call notes.