AI-proof interviews: how companies counter copilots

What is an AI-proof interview?

An AI-proof interview is one where hidden, live AI help either can't work or isn't needed. Companies have gone two ways. Some lock the process down with in-person rounds, written no-AI rules and proctored tests. Others open it up and let candidates use AI in plain view, then judge how well they use it.

Both directions answer the same problem. Remote interviews made it easy to read an answer off a second screen, and the interviewer saw only a face on a call. The changes below are what employers have said or published, each with a source and a date. Where a company has published nothing, this page says so and doesn't guess.

Which companies have changed their interviews, and how?

The documented changes fall into six kinds: in-person rounds, written no-AI rules, on-camera integrity statements, proctored online tests, open AI use and deeper follow-up questions. The table lists each with a company example and its source. All links were checked on October 22, 2026, and the dates shown are when each source was published or last updated.

The list is not complete, and a company's practice can differ by role and country. Before any round, ask the recruiter what's allowed. The current written policies for big employers are tracked on the employer AI interview policies page.

Why are companies bringing back in-person interviews?

Because an in-person round removes the second screen. Google is the clearest example. At a February 2025 town hall, employees asked to bring back onsite interviews, and Sundar Pichai said it was worth having "some fraction of the interviews being in person," according to an audio recording reviewed by CNBC.

By June 2025 that had become a plan. On the Lex Fridman Podcast, published June 5, 2025, Pichai said, "we are making sure we'll introduce at least one round of in-person interviews for people just to make sure the fundamentals are there." He also called the ability to use AI tools to write better code "an asset."

The trade-off is speed. Google's vice president of recruiting told the same town hall that the virtual process is about two weeks faster, CNBC reported. He said interviewers are told to probe candidates on their answers to check they understand what they're saying, which is the last row of the table.

CNBC also reported that Deloitte reinstated in-person interviews for its UK graduate program, citing a report from the previous September. Deloitte UK's own integrity guidance, checked October 22, 2026, says interviews "whether in person or virtual" must be completed without AI assistance.

What do written no-AI rules ask of candidates?

They ask you to agree, in advance, that the live interview is yours alone. Deloitte UK writes that "Interviews must be completed without AI assistance." Anthropic's candidate AI guidance, last updated July 10, 2025, says live interviews have "no AI assistance unless we indicate otherwise." Both encourage AI for preparation.

Amazon's guidelines, as quoted by Fortune on June 11, 2025, say "please do not use gen AI tools during your interview unless explicitly permitted." Fortune also quoted a Goldman Sachs campus recruiting email that "prohibits the use of any external sources, including ChatGPT or Google search engine."

Deloitte UK goes one step further. Before the online assessment, candidates record a short integrity statement on camera confirming they understand and agree to the guidelines. If you break the rule, the page says you'll be removed from the selection process. What that looks like in practice is covered in what happens if you're caught using AI.

What do proctored online tests check?

They watch the test session itself. HackerRank's Proctor Mode help page says the feature monitors that the candidate is present on webcam, flags attempts to leave full-screen mode or switch tabs, and analyzes screenshots for unauthorized tools such as external AI coding assistants. The employer chooses whether to switch it on.

HackerRank says Proctor Mode is available only for tests created after its July 2025 release. That is the vendor's description of what it monitors, and this page makes no claim about how often it catches anyone. The broader question of what interviewers and software can see is covered in can interviewers detect AI copilots.

If a test asks for webcam access, it will say so before you start. Read the instructions on the first screen, and if you need an accommodation, ask the recruiter before the test opens.

Why are some companies letting candidates use AI instead?

Because they decided policing it was the weaker option. Canva's engineering team wrote on June 11, 2025 that candidates were increasingly using AI, "sometimes covertly through tools specifically designed to avoid detection." Canva now expects backend, machine learning and frontend candidates to use tools such as Copilot, Cursor and Claude in technical interviews.

The Canva post explains the reasoning. Early tests found AI could quickly solve the old algorithm questions, so those questions no longer showed who could do the job. The new AI-assisted coding interview replaced the computer science fundamentals screen for backend and frontend roles and uses open-ended product problems.

What it scores has changed too. Canva lists judgment about when to use AI, breaking down vague requirements, and finding and fixing problems in AI-generated code. Its example question asks candidates to build a control system for aircraft takeoffs and landings at a busy airport.

Meta is moving the same way for some roles. 404 Media reported on July 29, 2025, from internal communications, that Meta told employees it will let some coding candidates use an AI assistant during interviews. The quoted reason was that it is "more representative of the developer environment that our future employees will work in." Meta has not published this as a public policy that we could find.

Why do identity checks come into it?

Because some fraud goes beyond AI answers to fake people. The FBI's public service announcement of June 28, 2022 described applicants for remote IT jobs using stolen identities and deepfakes in video interviews. It noted lip movements that did not match the audio. Employers responding to that risk now check who is on the call.

The security firm KnowBe4 wrote on its own blog in July 2024 that it hired a North Korean worker after four video interviews and a clean background check, because the stolen identity was real. Its advice to other employers included getting candidates on camera and asking them about the work they are doing, and checking that people are physically where they say.

For a real candidate, this mostly means more checks with no change in substance. You may be asked to keep your camera on and to answer detailed questions about your own past work. That last check is easy to pass when the projects are yours.

How should you prepare for an AI-proof interview?

Ask the recruiter what's allowed in each round, and get the answer in writing if you can. Then prepare to explain your reasoning out loud, because every change above, from in-person rounds to open AI use, ends up testing whether you understand your own answer. Practice with AI before the call; keep the call itself to you.

For an in-person or proctored round, rehearse your stories without notes. The STAR method gives each one a shape you can hold in your head. For an open-AI round like Canva's, practice with the tools the employer names, and practice saying why you accepted or rejected what the tool suggested.

After each round, write down the follow-up questions you got, because they show what the interviewer doubted. The free 10-minute debrief template has room for them. If there's another round coming, preparing for round two covers how to use those notes, and the framework for when AI in interviews is cheating sets out where the line sits.

Prept's line is analysis after the call, never during it, so none of these process changes affects how it works. With everyone's consent to the recording, the free tier debriefs one interview hour.