Does a flagged interview automatically get rejected?

No. Serin never auto-rejects on integrity signals. Every session gets a tiered rating and full evidence — the hiring team makes the call, the same way they would after noticing something odd in a live interview.

What actually gets recorded during a Serin interview?

The candidate's camera/mic session and a short pre-interview setup check, always. Screen share is recorded too whenever the organization has turned it on for that job — it's an opt-in setting per role, not something running silently in the background otherwise.

Can a candidate get flagged for something harmless, like glancing at a second monitor?

Isolated, low-severity signals like a brief window switch don't carry much weight on their own. The rating is a weighted combination of everything observed in a session, not a single strike.

Does this replace watching the interview yourself?

No, and it isn't meant to. It gives reviewers a starting point — what happened and when — so they can spend their attention on the sessions that actually need a closer look instead of reviewing every transcript blind.

The Integrity Problem Nobody Talks About in AI Interviews

Most conversations about AI interviews are about speed: fewer scheduling emails, faster first screens, more candidates seen per week. That's real, and it's why teams adopt them. But there's a question that gets asked far less often, usually only after something has already gone wrong: how do you know the person answering the questions is the person who applied — and that the answers are actually theirs?

Take away a recruiter sitting in the room, and an unattended AI interview is, structurally, one of the easiest things in a hiring process to game. A second tab with an AI assistant open. A friend feeding answers over a second device just out of frame. A script memorized from a forum thread that happens to match this exact question bank. None of that is hypothetical — it's the predictable outcome of putting a screening step behind a webcam with nobody watching.

The problem nobody puts in the demo

Vendor demos show the happy path: candidate joins, answers thoughtfully, gets scored, done. What they don't show is the failure mode that actually worries hiring teams:

A candidate reading a generated answer off a second screen, sentence by sentence
Someone other than the applicant taking the interview entirely
The same rehearsed answer appearing, nearly verbatim, across many candidates for the same role
A candidate stepping away mid-question while a script or another person supplies the response

If the interview itself is just a fixed list of questions, all four of these are close to undetectable. There's no follow-up to expose a shallow or borrowed answer, and there's no signal captured about what was actually happening around the candidate while they answered.

Why this hits AI interviews harder than human ones

A human interviewer picks up on this instinctively — a pause that's a beat too long, eyes that keep drifting to the same spot, an answer that doesn't sound like it came from the person's own experience. An AI interview doesn't get that instinct for free. If nobody deliberately builds it in, an AI interview is more exploitable than a human one, not less, because there's no one in the room to notice.

That's the gap most "AI interview" products leave open. Scoring the answer isn't the hard part. Knowing whether the answer deserves to be trusted is.

What actually solving this requires

Closing that gap takes more than a webcam recording nobody reviews. It needs three things working together:

Make it hard to cheat undetected — enforce the conditions a fair interview needs (a visible face, one person in frame, and, for roles where it matters, an unshared screen) instead of just hoping candidates comply.
Capture what happened, not just the transcript — so a reviewer isn't trusting a black-box score, they're looking at actual evidence.
Make the interview itself resistant to canned answers — the fewer answers a script or an AI assistant can supply word-for-word, the less integrity monitoring even needs to catch.

How Serin approaches it

This is where it stops being theoretical for us — it's what we've actually built into every Serin interview session:

Screen share, when a role calls for it. Screen recording isn't switched on for every interview by default — it's a setting an organization chooses per job. When it's on, candidates are told before the interview starts, sharing their screen becomes part of that session, and it's recorded; if sharing stops mid-interview, the candidate sees a countdown warning before the session ends. When it's off, screen share is never requested at all — there's no silent capture running underneath either way.

On-device face and person detection. Serin watches for no face in frame, more than one face, or an extra person visible alongside the candidate — running locally, with escalating warnings before anything drastic happens, so a candidate who briefly leans out of frame isn't punished the same way as someone who's clearly not alone.

Fullscreen enforcement. Leaving fullscreen during the interview is logged and flagged to the candidate in the moment, not discovered afterward as a mystery gap in the recording.

Quiet signals that don't interrupt the candidate. Tab switches, window or mouse focus leaving the interview, unusually large paste events, a second monitor appearing, or the camera and mic being swapped mid-session — none of these stop the interview, but all of them are logged, timestamped, and available to whoever reviews the session.

A single, weighted Integrity rating. Every session rolls all of this up into a High / Medium / Low rating sitting next to the transcript and scorecard — not a binary "flagged: yes/no." A candidate who glanced away once looks nothing like a candidate who had someone else answering from off-screen, and the rating is built to reflect that difference.

The agentic layer: why a pasted answer doesn't hold up

This is the part that's easy to miss, and it's the one we think matters most: Serin's interviewer isn't reading from a fixed script. As we covered in what an agentic interview actually is, it follows up on what the candidate actually said, not on the next line in a question list.

That matters for integrity in a very concrete way. A generated or memorized answer can survive a single scripted question. It rarely survives a specific follow-up about the candidate's own reasoning, the tradeoffs they didn't mention, or a detail that only someone who actually lived the example would know. In practice, this makes the interview itself part of the integrity system — not a separate layer bolted on top of it.

What this means for hiring teams

None of this hands a hiring team an automated verdict, and it isn't supposed to. It hands them the same thing a good in-person interviewer would have picked up on instinctively — context. A recruiter looking at a Medium-integrity session can see exactly what triggered it: one tab switch, thirty seconds long, nothing else. That's a very different conversation than a session with a fullscreen exit, a large paste event, and an answer that doesn't match the follow-up.

The bar we're holding ourselves to

Integrity monitoring that isn't explainable just moves the trust problem somewhere else — now you're trusting an opaque score instead of an opaque interview. We'd rather show our work: what was observed, when, and how much it actually weighs. The goal was never to catch as many candidates as possible. It's to make sure the hiring team can trust what they're looking at, and that an honest candidate never has to worry about a harmless glance away costing them a fair shot.