By the end of this lesson you'll be able to say exactly where AI help is allowed in a frontier-lab hiring process and where it isn't, and you'll have five concrete ways to use Claude for preparation that stay well inside the rules.
By the end of this lesson you'll be able to say exactly where AI help is allowed in a frontier-lab hiring process and where it isn't, and you'll have five concrete ways to use Claude for preparation that stay well inside the rules.
You're preparing for jobs where AI is the product, so it's tempting to assume the interviews are relaxed about AI. At Anthropic the opposite is true for most of the process, and the rules are written down. Getting them wrong doesn't make you look modern. It makes you look like someone who didn't read the instructions, in a job that is mostly reading instructions carefully. The good news: the rules are short, and they leave a lot of room to prepare well. Step through the stages above and read the "AI help" card for each.
Anthropic publishes guidance for candidates on using AI (last updated July 2025 when this lesson was checked on 2026-10-08). Everything in this section is Official, summarised in our own words; read the page itself before you apply, because it may change.
The guidance draws the line with examples. Using Claude to put real experience into better words is fine, such as working out how to quantify the impact of a project you led. Using Claude to invent experience is not, and neither is using it during an assessment that hasn't allowed it.
| Stage | What the guidance says | Label |
|---|---|---|
| Application (resume, written answers) | Write the first draft yourself. Then Claude can help you refine it, for example by comparing your resume against the job description to find what to highlight. | Official |
| Take-home assessments | Do them without Claude, unless Anthropic tells you AI is allowed for that specific assessment. | Official |
| Interview preparation | Encouraged. Use Claude to research the company, practise your answers and prepare questions to ask. | Official |
| Live interviews | No AI assistance unless Anthropic says otherwise. In the page's own words: "This is all you–no AI assistance unless we indicate otherwise." | Official |
| Accommodations | If you need one, tell your recruiter early. | Official |
Three expectations sit on top of the rules: use Claude as a collaborator, thoughtfully, to strengthen your application and preparation; be yourself (AI should refine your ideas, not replace them); and be transparent, to the same standard Anthropic says it applies to its own use of Claude.
The page also says how Anthropic uses Claude on its side of hiring (for things like drafting job descriptions and transcribing interviews), that it doesn't train Claude on candidate data, and that Claude doesn't make hiring decisions. That's Official too.
The rule in one line: before the assessment and around your own words, Claude is welcome. Inside an assessment or interview, assume no AI unless told otherwise in writing. If you're unsure about a round, ask your recruiter: "Is any AI tool allowed in the take-home?" is a normal question and costs you nothing.
This is the rule people bend without noticing. All of these break it:
The draft is the thinking. If Claude chose the structure and the points, the thinking isn't yours, however much you edit afterwards. A rough, short, slightly awkward draft that you wrote is the correct starting point. Then Claude can help.
Technically inside, but the wrong request. Refining your own first draft is allowed. But a full rewrite hands the voice and often the points to Claude, which is what "be yourself" warns against. Ask for critique, not a rewrite: "What's vague? Which sentence would work for any company? Where do I claim something without evidence?" Then make the changes yourself.
PROMPT TEMPLATES (paste after writing your own words first)
Mock round: "Act as an interviewer for a <round type> round. Ask me one original
question in the style of applied-AI interviews, wait for my answer,
then follow up on my weakest point."
Critique: "Here is my own first draft. Don't rewrite it. List what's vague, which
sentence would work for any company, and where I claim something
without evidence."
Quantify: "Here's a project I led and the numbers I have. What evidence would make
the impact clearer, and which numbers should I go and look up?"
The picture outside Anthropic is much thinner. For OpenAI, no official public candidate AI policy was found when this lesson was checked; one candidate account of the forward deployed engineering loop describes an AI-enabled coding screen, and third-party guides say the policy varies by team and round. That's Anecdotal: one account, and guides are not candidate reports. For Perplexity, no public policy was found (Unknown). If you're interviewing at OpenAI, the only reliable answer is the one your recruiter gives you for each round, so practise both ways: unaided and with an assistant.
BEFORE I OPEN CLAUDE IN AN APPLICATION OR INTERVIEW PROCESS
[ ] Is this an assessment or a live interview? -> If yes and nobody said AI is allowed: close it.
[ ] Am I starting from my own words? -> If not, write them first.
[ ] Would I tell the interviewer exactly how I used it? -> If not, don't do it.
If all three answers are good, use Claude as much as you like. It's one of the best practice partners you'll have.
Cheat card
- Anthropic (Official): own first drafts, then Claude may refine; no AI in take-homes or live interviews unless told otherwise; AI-assisted preparation is encouraged.
- Refine real experience, never invent it. "Draft it for me", "outline it for me" and "rewrite it" all move the thinking to Claude; ask for critique instead.
- Other labs: no official OpenAI policy found (the AI-enabled screen is one account); none found for Perplexity. Ask the recruiter, per round.
Assessment or live? -> no AI unless told. My words first? Would I disclose exactly how?Interview line "I wrote my answers myself and used Claude to critique them and to run mock rounds, and I asked my recruiter before each stage whether any AI tool was allowed."