Can you use ChatGPT or Claude on a take-home? The rules employers publish, and how to disclose

By FDE Interview Prep, a SourceStrongAI company

In this post12 sections

1. The rules, company by company
2. Which rule wins: the brief, the page, or the recruiter
3. Employers that say no AI unless they tell you otherwise
4. Take-homes that expect AI and ask you to show how you used it
5. When the brief says nothing: ask in writing
6. A disclosure note you can adapt
7. What to say when they ask “How did you use AI on this?”
8. Using AI so the reviewer still sees your judgment
9. Practice the parts no assistant can do for you
10. Questions people ask
11. Keep reading
12. More from the blog

The take-home brief is open in one tab and Claude or ChatGPT is open in the next, and the process has gone well so far. Using the assistant the wrong way is the one thing that can quietly end it: IBM’s candidate page, for one, says breaking its AI policy may get you disqualified. ^[1]Source 1Application processPublisherIBMSource typecompany job posting The take-home is one round in a longer loop, and the FDE interview guide covers the rest.

Short answer: only if the brief or the company’s candidate AI page allows it. If the brief says yes, use it and disclose it. If either says no, don’t. If both are silent, ask the recruiter in writing before you start.

Several employers publish their answer, and they disagree. Snowflake says no AI on take-home projects, Anthropic asks for take-homes without Claude unless it tells you otherwise, and Palantir’s Build to Apply postings encourage AI tools. ^{[2, 3, 4]}Source 2AI Cheat Sheet: How (and When) to Use AI in Your Snowflake InterviewPublisherSnowflakeSource typecompany hiring pageSource 3Guidance on Candidates' AI UsagePublisherAnthropicSource typecompany websiteSource 4Deployment Strategist, Build to Apply - US GovernmentPublisherPalantir Technologies (Lever)Source typecompany job posting

This post covers the published rules, what to do when there is none, a disclosure note you can copy, and what to say when they ask how you used AI.

The rules, company by company

Here is what each page or brief says. Rows marked “company-wide” come from a candidate page that covers every role; the rest are single take-home briefs or postings for FDE and related roles.

We read these pages in September 2026, and they change: OpenAI’s interview guide, for example, did not carry its current sentence on AI tools in archived copies from February to July 2026; the sentence appears in the August 2026 copy. ^{[9, 13]}Source 9OpenAI interview guide (Wayback Machine capture)PublisherOpenAI (archived by Internet Archive)Source typearchived company pageSource 13OpenAI interview guide (Wayback Machine capture)PublisherOpenAI (archived by Internet Archive)Source typearchived company page Check the live page on the day you start, not a table from a blog post, this one included.

Which rule wins: the brief, the page, or the recruiter

Several of these pages set a default and then say a specific assessment can override it. Anthropic’s version is the clearest: complete take-homes without Claude unless Anthropic indicates otherwise, and it says it will be clear when AI is allowed. ^[3]Source 3Guidance on Candidates' AI UsagePublisherAnthropicSource typecompany website

So the order of authority, as we read it, is:

1. The brief itself. If the take-home says what you may use, that is the rule for that take-home.
2. The company’s candidate AI page. If the brief is silent, this is the default.
3. A written answer from your recruiter. If both are silent, this is how you get a rule you can point to.

No AI on the take-home is not no AI at all. Anthropic encourages candidates to use Claude to prepare, including researching Anthropic and practicing answers, and Snowflake encourages AI for generating sample questions and brainstorming talking points. ^{[2, 3]}Source 2AI Cheat Sheet: How (and When) to Use AI in Your Snowflake InterviewPublisherSnowflakeSource typecompany hiring pageSource 3Guidance on Candidates' AI UsagePublisherAnthropicSource typecompany website Preparing with an assistant and doing the assessment with one are different things.

Company-wide pages still apply to you. They are not FDE-specific, but they cover you when you apply for a forward deployed role there.

Employers that say no AI unless they tell you otherwise

The table gives the defaults. Three details it can’t hold:

• Snowflake names take-home projects outright. It says candidates may not use AI to complete any written assessments, take-home projects or technical exercises, and the sentence has no “unless we say otherwise” clause. ^[2]Source 2AI Cheat Sheet: How (and When) to Use AI in Your Snowflake InterviewPublisherSnowflakeSource typecompany hiring page
• Accenture says it checks. Beyond its no-AI default for assessments such as coding challenges, it says it uses proctoring tools and AI-detection methods. ^[6]Source 6Redefining hiring with the power of responsible AIPublisherAccentureSource typecompany job posting
• Deloitte US’s rule covers interviews. Unless it says otherwise, you should complete interviews by yourself, without help from others or AI tools. ^[7]Source 7Using AI in the Deloitte recruitment processPublisherDeloitte USSource typecompany job posting The line is about interviews rather than take-homes, so for a Deloitte take-home, read the brief and ask.

The exception clause is why you read the brief first, and it can cut both ways at one company. Anthropic’s engineering blog describes a performance-engineering take-home on which candidates may use AI tools. ^[14]Source 14Designing AI resistant technical evaluationsPublisherAnthropicSource typecompany website That does not contradict its default. It is the “unless we indicate otherwise” case, written down.

A candidate-reported brief shows the strict end. One candidate reported, in a January 2026 repo that holds their submitted assessment, an AI Deployment Engineer take-home PDF headed “Take-Home Assignment: Voice AI Prompt Engineering” that says: “Don’t use ChatGPT to write this - we can tell.” ^[15]Source 15Take-Home Assignment: Voice AI Prompt Engineering (PDF)Publisherhobbes3 (GitHub)Source typecandidate’s take-home repository That is one brief, in one candidate’s repo, not a published company policy. But it shows that at least one brief says reviewers expect to recognize generated work.

Take-homes that expect AI and ask you to show how you used it

At the other end, some FDE take-homes are designed around AI tools, and they ask for evidence of how you used them.

• Palantir’s Build to Apply track. Palantir’s “Deployment Strategist, Build to Apply - US Government” postings ask applicants to build a Foundry and AIP project, submit a demo video, and say LLMs, AI and other technical tools are permitted and encouraged while you build. ^[16]Source 16Deployment Strategist, Build to Apply - US Government (Washington, D.C.) - Palantir on LeverPublisherPalantir Technologies (Lever)Source typecompany job posting The role is Deployment Strategist, not FDSE, but the build is the application; the Build to Apply post covers it in full.
• Coframe. Coframe’s public FDE take-home asks for a recorded video that covers the solution, including every AI tool you used and how it shaped the result. ^[11]Source 11Coframe Forward Deploy — Take-HomePublisherCoframe (GitHub)Source typecompany website The take-home walkthrough video post has a script for that recording.
• Namastex. Namastex’s public FDE and AI Engineer take-home expects AI coding tools and asks candidates to export all their AI conversations into an ai-logs/ folder, which is evaluated with the code. ^[12]Source 12Desafio Técnico — FDE / AI Engineer (Namastex)PublisherNamastex (GitHub)Source typecompany website
• ELITH. ELITH’s public FDE challenge asks for a GitHub repository with an FDE Decision Memo and an AI Usage Memo. ^[17]Source 17ELITH LOCK: FDE Challenge 2026PublisherELITH (elith-co-jp on GitHub)Source typecompany website

Briefs that candidates reproduced in their own repos point the same way. One candidate reported, in an August 2026 repo, a brief headed “Tavily - FDE - Take Home Assignment” that says: “We expect you’ll use AI tools — we want to see that you direct them well.” ^[18]Source 18Tavily - FDE - Take Home AssignmentPublisherazamkhan99 (GitHub)Source typecandidate’s take-home repository In a second repo, one candidate reported, in August 2026, an AI Agent Deployment Engineer take-home whose brief says: “You have to be able to explain how the code works, even if chatGPT wrote it”. ^[19]Source 19Hippocratic AI Coding AssignmentPublisherakshay-menta (GitHub)Source typecandidate’s take-home repository Each is one brief a candidate posted, not a company-wide rule.

The test in those lines is not whether you used AI, but whether you directed it well and understand what it produced. GitLab’s page says the same for technical assessments: transparent use of AI is welcome, with the expectation that you can explain your prompts and technical decisions. ^[10]Source 10Using AI in the interview processPublisherGitLabSource typecompany hiring page

For live rounds, which interviews allow AI tools covers the same employers, and AI-assisted coding interviews covers rounds where an assistant is expected. Pro’s lesson on AI rules by company and by stage sets every stage side by side.

When the brief says nothing: ask in writing

When the brief and the company page are both silent about AI, don’t guess and don’t infer a rule from a forum thread. Ask. OpenAI’s guide says it explains what is allowed in the preparation materials, and that candidates who are unsure should ask their recruiter before the interview. ^{[8, 9]}Source 8Interview guidePublisherOpenAISource typecompany hiring pageSource 9OpenAI interview guide (Wayback Machine capture)PublisherOpenAI (archived by Internet Archive)Source typearchived company page In our view, that is good advice at any company.

Our rule of thumb has three parts:

• Ask before you start, not the night before it is due. The answer may change how you plan the work.
• Ask in writing, by email or in the application portal, so the answer is on record.
• Ask about the gray areas by name. “AI” is vague. Inline completion in your editor, asking a chatbot to explain an error, and having it proofread your README are all AI use.

Here is a message you can adapt:

“Thanks for sending the take-home. Before I start, could you confirm the policy on AI tools for this assessment? Specifically: may I use an AI coding assistant (including inline completion in my editor), and may I use a chat assistant for looking things up or proofreading the write-up? If AI is allowed, is there a format you’d like for disclosing how I used it?”

If the answer is “no AI”, turn off inline completion for the whole task. If there is no answer by the time you need to start, work as if the answer is no, and say so in the README: “I did not hear back on AI tools, so I completed this without them.”

A disclosure note you can adapt

When AI is allowed, disclose it. GitLab’s page asks candidates to disclose which AI tools they used, and Coframe’s brief asks for every tool and how it shaped the result. ^{[10, 11]}Source 10Using AI in the interview processPublisherGitLabSource typecompany hiring pageSource 11Coframe Forward Deploy — Take-HomePublisherCoframe (GitHub)Source typecompany website Even where nobody asks, we’d write one: it costs you ten minutes and answers the question before anyone asks it.

This template is ours. Put it at the end of your README, under its own heading:

## How I used AI tools

Tools: [assistant name],
[editor completion]

Used for:
- Boilerplate: project setup, the CLI
  wrapper, test scaffolding
- Lookup: library API signatures
- Proofreading this README

My own work:
- The scope and the assumptions above
- The retry and error-handling design
- The eval set and the failure analysis

Checked: I read and ran every generated
line. Two suggestions were wrong: [what,
and how I caught each one].

Logs: [link or folder], if you want them.

A filled-in example

Here is an illustrative version for a brief like Coframe’s, which asks for an integration that syncs a spreadsheet to an API. The details are invented to show the level of specificity we’d aim for.

## How I used AI tools

Tools: a chat assistant,
editor inline completion

Used for:
- Project setup and test scaffolding
- The request client's boilerplate
- Proofreading this README

My own work:
- Scope: one-way sync only (see
  "Not built")
- Which failures to retry, and how
- The test fixtures and edge cases

Checked: I read and ran every generated
line. Two suggestions were wrong:
- It retried every failed request,
  including 400s. I limited retries to
  429 and 5xx with backoff: a 400 fails
  the same way every time.
- It read the sheet's header row as
  data. A two-row test fixture caught it.

Three details make it work:

• Name the split between the tool’s work and yours. “Used AI throughout” tells a reviewer nothing. “The retry design is mine; the test scaffolding is generated” tells them where to look for your judgment.
• Record what the tool got wrong. In our view, it is the strongest evidence that you checked the output rather than pasted it.
• Offer the logs, or include them if asked. Namastex’s brief evaluates AI conversations with the code. ^[12]Source 12Desafio Técnico — FDE / AI Engineer (Namastex)PublisherNamastex (GitHub)Source typecompany website A Tavily FDE brief that a candidate posted in August 2026 asks for a record of how the work was built, such as coding-agent chat history or session logs. ^[18]Source 18Tavily - FDE - Take Home AssignmentPublisherazamkhan99 (GitHub)Source typecandidate’s take-home repository If you may have to share logs, work as if someone will read them.

What to say when they ask “How did you use AI on this?”

If AI was allowed, expect the question in the follow-up call. Here is a spoken answer we’d give, built on the example above. The words are ours; swap in your own details:

“I used a chat assistant and editor completion for setup, test scaffolding and the request client’s boilerplate. The scope, the retry rules and the test cases are mine. I read and ran everything it produced, and it was wrong twice. The one that mattered: it retried bad-request errors, which will never succeed, so I limited retries to rate limits and server errors, with backoff. If I had another day, I’d add a dry-run mode so the customer can preview a sync before it writes anything.”

Four moves, in order: what the tool did, what you did, one specific catch, and what you’d do next. Practice it with what would you change in your take-home?, which is free and has a model answer, and with how you use AI coding tools, and where you don’t trust them (in Pro).

Using AI so the reviewer still sees your judgment

Our approach: let the tool type, and keep the decisions yours and visible.

Keep these for yourself. The scope and what you left out. The assumptions you made where the brief was silent. The test cases, especially the edge cases a customer would hit. The trade-offs and what you would change next. These are the parts you will defend live.

Mistakes we’d avoid, and the fix for each:

• Code you can’t explain. One candidate-posted brief says outright that you must be able to explain the code even if ChatGPT wrote it. ^[19]Source 19Hippocratic AI Coding AssignmentPublisherakshay-menta (GitHub)Source typecandidate’s take-home repository The fix is to read and run every generated line before it goes in. If you can’t say why it’s there, cut it or rewrite it.
• A README that sounds like a brochure. In our view, generic polished prose with no specifics is the first thing that looks generated. The fix is to write the assumptions and trade-offs yourself, in plain sentences with real details.
• Far more code than the time box allows. The Tavily brief a candidate posted puts it plainly: “We’d rather see a small thing done well than a large thing done loosely.” ^[18]Source 18Tavily - FDE - Take Home AssignmentPublisherazamkhan99 (GitHub)Source typecandidate’s take-home repository The fix is a small thing done well, with a clear list of what you chose not to build.
• An undisclosed tool that shows up in the walkthrough. The fix is the disclosure note above, written before you submit.

Practice the parts no assistant can do for you

Whatever the rule, you will be asked to explain how you scoped the problem, what you assumed, and why. To rehearse that against someone who answers back, the free practice case takes a sign-in and nothing else.

If your take-home is due this week, Pro has the lesson that sets out every company’s AI rule by stage, the AI-use and take-home questions with model answers, and the lessons on scoping a problem before you build. The question bank holds 180 interview questions, each with a framework and the follow-ups to expect. Pro starts with a 7-day free trial, billed monthly or annually after that. Start the free trial.

Questions people ask

Is using ChatGPT on a take-home cheating?

It depends on the employer’s rule. Snowflake says candidates may not use AI on take-home projects, and Anthropic asks candidates to complete take-homes without Claude unless it says otherwise. Other briefs expect AI tools: Palantir’s Build to Apply postings encourage LLMs while you build. Read the brief and the company’s candidate AI page before you start.^{[2, 3, 4]}Source 2AI Cheat Sheet: How (and When) to Use AI in Your Snowflake InterviewPublisherSnowflakeSource typecompany hiring pageSource 3Guidance on Candidates' AI UsagePublisherAnthropicSource typecompany websiteSource 4Deployment Strategist, Build to Apply - US GovernmentPublisherPalantir Technologies (Lever)Source typecompany job posting

What should I do if the take-home instructions don’t mention AI?

Ask the recruiter in writing before you start, and keep the reply. OpenAI’s interview guide, for example, says expectations for AI tools vary by interview and tells candidates who are unsure to ask their recruiter.^{[8, 9]}Source 8Interview guidePublisherOpenAISource typecompany hiring pageSource 9OpenAI interview guide (Wayback Machine capture)PublisherOpenAI (archived by Internet Archive)Source typearchived company page

Should I disclose AI use on a take-home?

When AI is allowed, disclosing it is the safe default, and some employers ask for it. GitLab’s candidate AI page asks candidates to disclose which AI tools they used, and Coframe’s public FDE take-home asks candidates to list every AI tool they used and how it shaped the result.^{[10, 11]}Source 10Using AI in the interview processPublisherGitLabSource typecompany hiring pageSource 11Coframe Forward Deploy — Take-HomePublisherCoframe (GitHub)Source typecompany website

Keep reading

Guides

• The forward deployed engineer interview, round by round

Lessons

• AI assistance rules, by company and by stage
• The evidence map: what employers publish, what candidates report, how to read both

Questions

• How do you use AI coding tools in your daily work, and where do you not trust them?
• Extend a feature in a Python codebase you have never seen, using an AI assistant, and explain every change it made.
• Looking at your take-home submission, what would you change with another day, and what did you leave out on purpose?

More from the blog

1. What the evidence says

   Which companies allow AI tools in interviews? The published rules, by stage

   Some employers expect you to use an AI assistant in interviews; others may disqualify you if you do. The published rules by company and stage, dated.

   10 min readPublished September 2026

2. What the evidence says

   Do FDE interviews have LeetCode? What employers publish and candidates report, company by company

   Some FDE loops include LeetCode-style rounds and some have dropped them. What employers publish, what candidates report, and how to split your prep.

   11 min readPublished September 2026

3. What the evidence says

   FDE interview pass rates, round weights and timelines: what no one publishes

   No FDE employer we checked publishes a pass rate, round weight or timeline. Where the circulating figures come from, and how to check any claim yourself.

   9 min readPublished September 2026

More in What the evidence says