The Take-Home Says No AI: What Happens If You Use It Anyway

Expect a Follow-Up Conversation, Not a Detector

Detection is not the mechanism to plan around. An ICSE 2025 study tested five general AI-content detectors plus one built for code, on benchmark code; accuracy mostly fell below 0.6, which the researchers called ineffective, and even their purpose-built replacement reached only an F1 in the low 80s, nowhere near a standard that could support rejecting anyone ¹. Text detectors fare no better: a study of fourteen AI-text detection tools, two of them commercial systems, found none accurate or reliable ⁶.

What's actually changing is the task. One engineering team that has run the same take-home since early 2024 found its own newest models progressively beating the exercise, and responded by redesigning it three times rather than trying to police submissions ²; in a 2025 survey of UK early-careers employers, 79% said they were redesigning or reviewing their recruitment process in response to AI ⁵. What graders are told about polished take-homes points the same direction: grade the candidate's decisions and ask about them, because a polished artifact on its own tells them almost nothing.

Prepare for that conversation rather than for a scan you'll likely never encounter. Know why you made each significant choice in the submission, what you tried that didn't work, and where you'd change the approach given more time. A candidate who can answer those questions clearly is ready for that follow-up, regardless of what wrote the first draft of any given function.

What a Stated No-AI Rule Actually Means

If the instructions in front of you say no AI, treat the rule itself as a real part of what's being scored, and take it as seriously as any other requirement in the brief. One large public employer, the UK Civil Service, states the consequence directly: applications may be rejected where AI is used inappropriately, and the one absolute line it draws is no AI during any live assessment at any stage ³.

Employers are working through the mirror version of this question right now. Whether a candidate who clearly used AI on a take-home should be disqualified turns, on their side, on whether a written rule existed before the task started and was held the same way for every candidate who sat that same assessment. A rule you were told about in advance is exactly the kind that gets enforced when it's broken.

The same principle protects you if the accusation runs the other way and nothing was actually wrong. A rule invented after the fact, applied only once someone reads a submission and feels suspicious about it, is not a rule; it's a reaction dressed up as a policy, and it's worth naming as the difference it is if you're ever on the receiving end of one.

If the brief says nothing where you expected a rule, don't read the silence as permission. Ask before you start, in one line, the same way you'd ask about scope or length. A written answer protects you in both directions: it licenses the tools you planned to use, and it's the record you point at if anyone questions the submission later.

Ask Before the Deadline When the Rule Doesn't Fit

Some no-AI rules don't match the actual task: a take-home expecting a production-quality deliverable in three hours while banning the tools most engineers use daily is a rule in tension with itself. When that happens, ask before the deadline rather than deciding silently which instruction to honor.

A short, specific question to the recruiter, does the no-AI rule cover research and documentation or only the code itself, usually gets answered, and it's a far better position than guessing wrong under a deadline pressure you created for yourself by staying quiet. Silence reads as a choice once the deadline has passed, even when it was really indecision.

Take the rule seriously regardless of how that ask goes, because the consequence of guessing wrong can outlast the interview stage entirely. Employers asking whether an offer can be pulled back after the fact over a broken take-home rule are told the answer is usually yes in the US, where offers are presumed at-will before day one, provided the rule was in the brief before the task and held the same way on every candidate. That exposure is the reason a rule you personally disagree with is still worth asking about rather than quietly ignoring it entirely.

What Makes a Submission Defensible

Where AI is allowed or the instructions say nothing, the version that wins is the one you can defend, which usually means visible reasoning left in rather than maximum polish sanded over the seams of how you actually got there.

One AI company publishes a stage-by-stage rule for candidates rather than a blanket ban: draft the written application yourself and refine it with AI, use AI freely to prepare for interviews, and skip it on take-homes and live interviews unless told otherwise for that specific assignment ⁴. Reading the instructions for the exercise actually in front of you, not a general policy remembered from somewhere else, is the habit worth carrying between employers, since the specific brief can override the general one without much warning.

In the Institute of Student Employers' 2025 survey of UK early-careers employers, half reported no problem with candidates using AI in the process, only one in ten had banned it or blocked it technically, and nearly half had given applicants no guidance on when it is or isn't appropriate ⁵. That gap is why asking is a normal move rather than a nervous one: for nearly half of these employers, no written answer exists until you request one.

If nothing is written anywhere, a short note in your submission saying what you used and where costs you almost nothing and closes the gap yourself, rather than waiting for a reviewer to wonder about it in silence and decide the answer without you in the room.

See what gets scored >