Use Their AI Account If They Provide One, Ask If They Don't

Which of the Three Setups Are You In?

Three setups are worth telling apart. The employer hands you a tool inside their own environment, the brief tells you to bring your own, or nobody says anything at all. Each has a different right move, and an easy mistake is treating the third case as though it were the second.

The provided case is the easiest to get right, because the environment they built can also be the record they review. CodeSignal's AI-assisted assessments, launched 28 May 2025, put an assistant inside the assessment IDE in two modes the employer picks between ¹. One is a full co-pilot mode, for tasks the release says would otherwise be unsolvable in the time allowed. The other is limited to platform navigation, syntax and documentation.

Which of the two you meet is the employer's setting, not your choice.

Your own account is a different arrangement. An employer hiring for how you work inside a stack you already know has a reason to let you bring it, and a brief that means this should say so. Keep the session if you do. Being able to show the working afterwards is most of the advantage of having done it in your own tools.

The third case is common.

When nothing is stated, the honest reading is that the employer either has not decided or has decided and not told you. Neither is your fault, and neither gets fixed by guessing well. A line saying AI is allowed has answered a different question from the one you are asking, because allowed says nothing about whose account, which stage, or what happens to the session afterwards. Those are three questions and they can have three different answers.

Employers get their own version of this question.

Somebody has already written down what the brief should have said. What employers are told about allowing AI on a take-home works through the allow-or-ban decision from the other chair. It tells them to put the rule in the brief in three lines: what is allowed, what must be disclosed, what gets graded.

A brief missing those lines has quietly handed the decision to you, and knowing what was supposed to be there makes the gap easy to name when you ask.

What Does the Employer Read Afterwards?

In a provided environment, the session can be part of what gets read. CodeSignal says hiring leaders receive transcripts of all AI interactions alongside session replays, which the release frames as a transparent view of how candidates used the assistant ¹. It does not say how those transcripts are weighed. It says only that they exist, which is reason enough to write your prompts as if a person will open them.

Live formats work the same way with a person in the room. Karat launched NextGen Interviews on 10 December 2025, a format in which engineers work on multi-file projects with an integrated AI assistant while a Karat Interview Engineer probes reasoning, trade-offs and judgment in real time ². That is a vendor describing one product line, not evidence that every Karat interview now allows AI. Which format you meet is still the employer's decision.

Stepping outside the environment removes the part they may have planned to read.

Platforms also publish what an assessment environment can log, and reading that list once is better than wondering about it all afternoon. HackerRank's Screen page lists tab proctoring, copy and paste tracking, multiple-monitor detection and full session replay among its signals, alongside a plagiarism model and an AI proctor ³. No accuracy figure is published for any of it. Which signals are switched on is the employer's choice, and the page does not say which are on by default.

The framing on that page is worth more than the list. HackerRank states the rule per test: if a test is designed to allow AI, it is fair for everyone to use it, and if it explicitly instructs otherwise, using AI will be flagged as suspicious activity ³. A flag is a note an employer sees, not a verdict.

The same platform can allow AI on one test and instruct against it on the next.

The employer-side piece on running a coding interview with AI open names what the interviewer is meant to score. Three of its four columns are what got tested against something outside the chat, what got refused and on what grounds, and what the candidate kept for themselves. How much AI you used stays off that sheet. On that sheet, what you did with it is the part being read.

Ask Before the Clock Starts

If the brief says nothing, ask, and ask before you start rather than afterwards. The silence is common. In the Institute of Student Employers' 2025 recruitment survey, 45% of employers had given applicants no guidance at all on when AI use is appropriate ⁴. 49% said they had no problem with candidates using AI during recruitment, and the survey ties that acceptance largely to drafting cover letters, CVs and online application answers ⁴.

Those are UK early-careers employers answering a membership body's own survey, and the full report is member-gated, so the sample size and the question wording are not open to check. An attitude reported to a survey is also not a policy you can rely on. The 45% with nothing published is exactly the group whose position stays invisible to you at application time.

One employer's published rule shows what a clear one looks like. Anthropic's candidate guidance, dated 10 July 2025, answers per stage rather than yes or no ⁵. Candidates are asked to draft the application themselves and use Claude only to refine it, encouraged to use it in interview preparation, and told to complete take-home assessments and live interviews without it unless Anthropic indicates otherwise ⁵.

That is one company's policy about its own hiring, not an industry norm. The page names no penalty for breaking any part of it, and Anthropic says it plans to review and update the guidance.

A rule you can read beats a permissive attitude you cannot see.

Borrow the shape of it for your own question.

• Is AI allowed at this stage specifically, and is the take-home separate?
• Do you provide an account, or should I use my own?
• Is anything from the session kept, and does the reviewer see my prompts?

One habit outlasts whichever policy you meet. Karat's candidate page of 9 February 2023 allowed looking up documentation during the coding portion, including with ChatGPT, provided you told your Interview Engineer first, and did not allow it to generate an approach or a solution ⁶. Karat has since launched an AI-enabled format, the kind of update the page itself said might come.

Telling the interviewer before you reach for a tool has not been overtaken.

Should You Bring Your Own Editor?

Bring your own editor when the brief lets you work locally, and stay in theirs when the assessment runs in their browser. A local project with a deadline is yours to set up however you like. A timed session inside a hosted IDE can be the record itself, and leaving it takes the record with you.

There is a real cost to working in an unfamiliar editor, and saying so is fair. Ask whether you can build locally and paste in, or whether the environment takes your keybindings, before the session opens rather than while the timer runs. Some of the friction people blame on nerves afterwards is a shortcut that was not where their hands expected it.

Raise the money question early too.

If the exercise assumes an assistant you do not pay for, a subscription has quietly become part of the entry price, and that is worth naming. Raising it a week ahead is easier on both sides than a thin submission nobody can explain. Ask what they expect you to use and whether an account comes with the exercise.

That question belongs in the same message as the others.

What they are watching for when AI is allowed in a coding round covers the live version of this, once the assistant is open and the interviewer starts asking follow-up questions.

One message, sent a day early, asks about all three cases.

See what gets scored >