Somewhere in the last two years, the question changed. Hiring teams used to ask whether candidates were using AI tools during interviews. Now the honest question is how many, and most talent surveys put the answer somewhere above half.
That number makes some recruiters uncomfortable. But the companies hiring well in 2026 have mostly stopped treating AI assistance as a cheating problem and started treating it as a design problem. The difference matters, and it is quietly reshaping how interviews work.
The Detection Arms Race Nobody Is Winning
The first response to AI-assisted interviewing was surveillance. Eye-tracking software, browser lockdowns, keystroke analysis, and tools that claim to detect when a candidate is reading generated text.
The results have been underwhelming. Detection tools produce false positives that punish nervous candidates, non-native speakers, and people who simply look away from the camera while thinking. Meanwhile, determined candidates route around every new control within weeks of its release.
There is also a cost that shows up later in the funnel. Candidates talk, and heavy-handed proctoring signals distrust before someone has even joined the company. Several large employers have quietly walked back their strictest monitoring policies after watching offer acceptance rates dip.
Why Policing Fails the Logic Test
Set aside the technical problems and the logic still wobbles. Most companies now encourage employees to use AI tools on the job. Engineering teams ship code with AI pair programmers. Marketing teams draft campaigns with them. Support teams resolve tickets with them.
An interview that bans the tools someone will use daily is not measuring job performance. It is measuring performance in an artificial environment that disappears the moment the candidate is hired.
That does not mean nothing matters anymore. Judgment, verification, and communication under pressure still separate strong hires from weak ones. The point is that the interview needs to measure the skill that survives the tool, not the skill the tool replaced.
What Leading Teams Are Doing Instead
The shift showing up across hiring organizations follows a few clear patterns.
Some teams have made AI use explicit. The candidate is told which tools are allowed, sometimes even given a prompt window, and evaluated on how they direct, question, and correct the output. A candidate who accepts a wrong AI answer without checking it reveals more in five minutes than an hour of trivia questions ever did.
Others have moved weight toward live collaboration. Pairing sessions, working through ambiguous problems with the interviewer, and debugging exercises where the starting code came from a model. These formats are hard to fake because the signal is the conversation itself, not the artifact produced at the end.
A third group has rebalanced the funnel. Less emphasis on take-home output that could be generated in an afternoon, more emphasis on structured discussion of past work, with follow-up questions that go three levels deep. Fabricated experience tends to collapse around the second follow-up.
The New Skill Is Verification
Across all of these formats, one capability keeps rising in importance: knowing when the machine is wrong.
AI output is fluent, confident, and sometimes incorrect. The employees who create real value with these tools are the ones who catch the errors, ask for sources, test the edge cases, and know when to throw the draft away. That is an observable skill, and interviews can be built to surface it.
Expect to see more interview questions that hand candidates a plausible but flawed AI answer and ask what they would do with it. It is a cheap format to run, and it maps directly onto daily work.
What This Means for Candidates
If you are job hunting, the practical advice has flipped. Hiding AI fluency is now a worse strategy than demonstrating it well.
Be ready to talk about how you use these tools, where they fail you, and how you check their work. Bring examples. A specific story about catching a model's mistake is a stronger signal than any claim about prompt writing skills.
And when a company does run a no-tools interview segment, respect it. Plenty of roles still require unassisted reasoning, and the fastest way to lose an offer is to break explicit rules.
The Bigger Shift
Interviews have always lagged the workplace by a few years. Whiteboard coding outlived the whiteboard. Brainteasers outlived the evidence against them.
The AI transition is compressing that lag. Companies that redesign now are getting cleaner signal, better candidate reviews, and fewer awkward integrity disputes. The ones still investing in detection are spending money to measure the wrong thing with increasing precision.
The interview of 2026 is not AI-free, and it should not pretend to be. The winning teams are asking a better question: not whether this candidate used AI, but whether they would want this candidate using AI on their team.