To get into Anthropic, you first have to pass five rounds of rigorous interviews and skills tests, during which you are not allowed to use AI tools, and you have to sign an NDA (non-disclosure agreement) before meeting with any employee other than the hiring manager. But the final hurdle that eliminates the most people is the culture interview.
Bloomberg Businessweek interviewed several anonymous candidates, former recruiters, and career coaches and reported that Anthropic's cultural interviews are known for their intensity. Interviewers can come from any department, and a low score can disqualify a candidate. The whole process has been described by many as "more like psychological counseling than an interview."
CEO Dario Amodei said he spends about a third to 40 percent of his time ensuring the company culture stays on track, a percentage almost unheard of among tech executives.
What do cultural interviews test? Not beliefs, but the ability to think independently.
According to Kevin Landucci, a career coach at Exponent, interviewed by Bloomberg, the core question in anthropic culture interviews is: "How did you feel about this dilemma, how did you respond, and how do you see it now?" Whether you're applying for a research scientist, accountant, or payroll specialist position, almost everyone has to pass this test.
Landucci's advice is to demonstrate "unease" by choosing an ethical choice that makes you hesitate but doesn't shake the company's foundation, such as a questionable decision involving user data. Anthropic President Daniela Amodei explained the selection logic in a podcast: "We're not looking for a specific belief system, but rather the ability to say, 'I feel this is right, it may not be popular, but I stand by it.'"
The concept of "intellectual independence" is also a key signal. Landucci points out that Anthropic even hopes candidates will be skeptical of the company itself and its approach to pursuing its mission, rather than accepting it wholesale. This is quite different from the approach of most companies that emphasize cultural fit: the latter wants you to "integrate," while the former wants you to "dare to challenge."
A former recruiter who joined the lab last year, who requested anonymity, revealed that part of his motivation for speaking to Bloomberg was to push for greater transparency at the lab.
Why are people scrambling to hire? Salaries, retention rates, and the talent pool.
The other side of this rigorous selection process is an exceptional compensation structure. Anthropic's salary packages typically exceed $250,000, with key positions offering as much as $850,000; coupled with stock awards, early employees who survived until the IPO could potentially receive hundreds of millions of dollars in present value.
Financial advisor Nicholas Garcia described how many early employees of top AI companies like Anthropic would no longer need to work once the company goes public.
Anxiety has spread to the job preparation industry. Data from the Interviewing.io platform shows that users who successfully entered Anthropic or OpenAI spent an average of about $4,600 on preparation, with mock interviews costing $170 to $550 or more per hour. Lerner, the platform's founder, bluntly stated, "Spending a few thousand dollars and getting a $200,000 salary increase is a worthwhile deal."
Retention rates validate the company's attractiveness: SignalFire analysis last year showed that Anthropic's two-year retention rate was 80%, the highest in the industry. Even more compelling is the flow of talent: engineers are 8 times more likely to move from OpenAI to Anthropic than the other way around; and about 11 times more likely to move from DeepMind.
This month, Andrej Karpathy, a founding member of OpenAI and a former Tesla executive, also joined the Anthropic research team, becoming the latest example of this talent drain. Even Peter Bailis, the CTO of Workday, willingly gave up his title to take on a "technical" position at Anthropic.
Related reports
Anthropic CEO Dario Amodei: In another 6-12 months, Chinese open-source AI models will catch up.
