Bloomberg reports that generative AI giant Anthropic has implemented an industry-controversial and extremely difficult interview process, banning job seekers from cheating with AI tools and introducing a so-called "therapy-style" culture interview.
As the company's employee base has rapidly expanded over the past six months, CEO Dario Amodei revealed he spends nearly a third to 40% of his time on culture building. This culture interview has a one-veto power, and its core purpose is to screen for candidates who are highly aligned with the safety mission and capable of independent thinking.
Candidates must go through up to five rounds of rigorous screening, including signing a non-disclosure agreement (NDA). During the interview, using large language models is strictly prohibited unless explicitly authorized. In the culture interview, which has a very high elimination rate, interviewers apply rapid-fire questioning to pressure candidates continuously to extract evaluation signals, even cutting them off mid-answer when no new information is being added. Some candidates revealed that interviewers show no facial expressions or feedback when asking questions, creating immense psychological pressure.
The core of the interview is to assess candidates' values and their understanding of the frontier of existential risks from AI. Interviewers pose classic ethical dilemmas, probing candidates' emotional reactions, coping actions, and post-mortem reflections during conflict moments, aiming to evaluate genuine moral struggle. Additionally, candidates are asked: "What unusual beliefs have you held onto, and how did you defend them in difficult situations?"
President Daniela Amodei noted that the interview does not require candidates to adhere to a fixed set of dogmas, but rather tests their ability to have rational discussions with people who hold different opinions. The company wants people who dare to question decisions and maintain independent thinking, not yes-men who blindly follow.