A. Osei · candidate
Patent-pending · Closing the gap in remote hiring
Meeting Platform
Engineered for Interviews.
Interviews still run on tools built for meetings. That gap is where hidden AI help, stand-ins and deepfakes get through. Capifiq closes it: video, a shared code pad, a whiteboard and live captions in one room, with a timestamped record of what actually happened.
A simulated interview in progress, showing the shared code pad, whiteboard, the candidate’s connected displays, and a timeline of integrity observations. Every control below can be operated. Skip the interview preview
capifiq.
Full observation
Candidate18 ms · 0.0% loss
You24 ms · 0.1% loss
Video720p · stable
You
4 in the room Interview integrity active
rate_limiter.py
Candidate has the pen
1
# regional token bucket — shared counter in redis2
import time3
4
class RateLimiter:5
async def allow(self, key):6
now = time.monotonic()7
bucket = await self.r.hgetall(f"rl:{key}")8
tokens = min(self.burst, float(bucket.get("t", self.burst)))9
last = float(bucket.get("ts", now))10
11
tokens = min(self.burst, tokens + (now - last) * self.rate)12
if tokens < 1:13
return False14
15
await self.r.hset(f"rl:{key}", mapping={"t": tokens - 1, "ts": now})16
return Truecandidate has the pen
Displays open one at a time. Opening one is recorded in the session timeline.
Display 2recorded
Two observations, from different subsystemsScreen area not readable · text appeared without matching typing
Sessionfull observation
00:0015:0030:0045:0060:00
14:32 – 14:33 · two observations, different subsystems
Earlier
Interview preview paused
Human interviewerNo fit scoreNo rankingNo emotion inferenceNo automated rejection
Why a purpose-built platform
An interview is not a meeting.
Meeting tools assume everyone on the call wants the same thing. In an interview that isn't true. Someone has a job on the line, and a whole industry is selling them ways to win it.
Add-ons look in the wrong place
A browser plugin sees the browser. A bot in the call sees the call. The tools that beat interviews sit outside both, on the desktop, where neither one can reach.
So we moved the interview
Run it in Capifiq and the app your candidate is already in is the one keeping things honest. Nothing to bolt on, and nothing falling through the gap between two tools.
One interview, one record
Video, transcript, timeline and findings all come out of the same session. No exporting from three places to piece together what actually happened.
Only your interviews move. Your ATS, your calendar and your assessment tools stay exactly where they are. See what Capifiq connects to →
In the room
Built for the interview, not adapted to it.
When you build a room just for interviews, you can put the right things in it, and leave the wrong things out.
Write code together
A shared pad you can both type in. No screen sharing, no "can you see my editor?", no scratch file that disappears the moment the call ends.
Sketch the system
A whiteboard for the design round. It's captured in the recording, so the diagram is still there when you sit down to write up your notes.
The pad is never a test
Nothing written on it gets graded or scored, and we never read anything into how fast someone types. It's a place to think out loud. That's all it is.
Captions, on by default
Everyone follows the conversation, including candidates who need them to. And a caption stays a caption: no tone analysis, no keywords flagged.
Bring the whole panel
Invite co-interviewers, observers and guests by email. No accounts to provision, no licences to buy, no scramble for access at 8:59.
Notes that stay yours
Type freely during the interview. Nothing saves until you press save, and your notes never reach the integrity report. Whoever reviews the evidence stays independent of your opinion.
Complete coverage
One platform for the whole interview environment.
From the person on camera to the device, the environment, and any live assistance. Capifiq verifies what a screen share can't show, and seals each finding as evidence. These are common categories, not a fixed list. Select one to see how it's abused and what Capifiq reveals.
Deterministic Proven, on-device evidence, every category below is surfaced as a timestamped, cryptographically sealed event.
Hidden AI Assistants
Desktop & web AI tools
Remote Control Sessions
AnyDesk, TeamViewer, Chrome Remote
Virtual Machines
VMware, VirtualBox & emulators
Virtual Cameras
Blocked at join: OBS, ManyCam
Browser-Based Assistants
Extensions & web apps
Display Tampering
Screen overlays & manipulation
Identity & Proxy Candidates
Stand-ins & coached impostors
Deepfakes & Synthetic Media
AI face & cloned voice
Location & VPN Masking
Hidden or mismatched location
Unauthorized Input
Injected & alternate input devices
How it can be used to cheat
Answers appear beside the interview
A hidden desktop or web assistant can listen to the conversation and surface generated answers while the interviewer sees what appears to be a normal video call.
Analytical signals · model-derived, human-reviewed
Media-authenticity signals, labelled separately.
The evidence above is a record of things directly observed. Media authenticity is a different kind of claim. It is model-derived and probabilistic, so Capifiq labels it separately, never calls it proven, and routes it to a person. It describes the feed, never the candidate: nothing here is an inference about who they are or how they behaved.
Synthetic-media & deepfake cues
Audio-visual consistency of the feed
Cloned-voice cues
Every signal here is a statement about the video and audio feed. None of them is an accusation, a score, or a decision, a person reviews the record and decides what it means.
How it works
From scheduled interview to a sealed integrity report.
Schedule the interview in Capifiq, invite the candidate, monitor integrity live, review any findings, and finish with a sealed report.
Schedule the interview
Add the role, the panel, and the time. Capifiq schedules the interview and creates the protected session it runs in. There is no external meeting link to paste.
Interview
Software Engineer: Final round
Final roundSession ready
Schedule interview
Candidate receives their join link
Capifiq emails the candidate clear instructions and a secure link to join the interview in the Capifiq client. The integrity check comes with it. There is nothing separate to install.
Your interview integrity checkSent automatically by Capifiq
Join the interview→
Monitor integrity live
While the interview is running, the interviewer can confirm session integrity and watch verified events appear on the live dashboard.
Interview in progressLive
Candidate joined the interview
Environment verified
AI overlay detected
Review and act on findings
When Capifiq surfaces an event, the interviewer sees it in the moment and decides how to proceed with the conversation. Dispositioning the record afterwards is a separate authorised reviewer's job.
Hidden AI overlay detectedDeterministic integrity finding · sealed
Detected at10:24:31 AM
Evidence statusSecured
Report created and sealed
At the end of the interview, Capifiq creates an integrity report with the session timeline, findings, timestamps, and sealed evidence.
Interview Integrity Report✓ Sealed
SessionFinal round
Findings1 verified
Evidence hash
a3f6b7c9e2d4…✓
Only the interview moves. Your ATS, calendar and assessment stack stay where they are.
The damage is already measurable
The hiring threat is accelerating.
AI-assisted interview fraud is not a future scenario. Its prevalence, sophistication, and downstream cost are already moving in the wrong direction.
JulSep
Fabric reported that 38.5% of candidates were flagged across its analysis of 19,368 interviews.
Fabric · 19,368 interviews
Every interview run without integrity verification is a bet that the next hire is not one of them.Verify the interview before the risk becomes payroll, access, rework, and replacement cost.
AI interview assistance is now an industry
You're not defending against three apps.
You're defending against an ecosystem.
The tools that coach candidates through your interview are venture-funded, growing fast, and multiplying, with open-source forks and private builds shipping continuously. A denylist can't keep up, which is exactly why Capifiq detects the behavior, not the name.
$20.3M
raised by Cluely, led by a16z: a tool built, in its own words, to “cheat on everything.”
TechCrunch · Jun 2025
10M+
users claimed by Final Round AI: one interview-copilot tool among dozens.
Vendor-reported
1.53M+
users claimed by Parakeet AI, an on-screen answer overlay.
Vendor-reported
Figures vendor-reported except where sourced. Updated July 2026.
The blind spot
Screen sharing shows the interview.
Capifiq verifies the environment.
Hidden tools. Remote access. Virtual environments. You can't see them. We can.
How candidates cheatSee how hidden AI overlays work in real interviews, then secure your next interview with Capifiq.
Same interview
Capifiq verifies
1
2
3
What you see
Interviewer's screen
Visible Standard video interview
Clean interview surface
What's really happening
Candidate's screen
Interview
System Design Round
RoleSWE II
InterviewerSarah
StatusLive
Typing answer…
Hidden AI overlay
Unauthorized assistance detected
What Capifiq detects
Live evidence inside the Capifiq dashboard
Interview integrityDetection review
LiveAlex JohnsonSoftware Engineer, Final Round
Live detection feed3 events
Hidden AI overlay detected
Unauthorized desktop assistant active
•
Browser extension activeInterview helper extension
•
Remote session checkedNo active connection
The near-webcam second-device attack
The cheating tool may be outside the computer.
A candidate can mount a phone on the monitor bezel, let it hear the interview through its own microphone, and read generated answers while appearing to look naturally toward the webcam.
Off-host assistance pathPhone microphone → AI answer → candidate
!The phone is not paired, mirrored, or connected to the candidate's computer. To the operating system, there is no second device to find.
The AI smart-glasses attack
Or the answer is on the candidate's face.
AI smart glasses can show generated answers inside the lens, in the candidate's line of sight: no screen to share, no window to hide, and nothing installed on the interview computer.
On-face assistance pathInterview audio → AI answer → in-lens display
!The glasses run on their own battery and network. Nothing is installed on, paired with, or shown on the candidate's computer.
Detection covers the device. The questions cover the rest.
A phone on the bezel, glasses on the face, an earpiece in the ear: none of it touches the candidate's computer, so no software can see it. Everything that runs on the machine is caught and sealed as evidence; everything that doesn't runs into questions built so that help of that kind has little to offer.
See the Question Engine →
The full picture
Detect what's on the device. Deter what isn't.
Interview integrity has two halves, and both are needed. Capifiq surfaces covered integrity-relevant activity on the candidate's machine, and outpaces the off-device help no software can see.
Detect · on-device
Any assistance on the machine
Hidden AI overlays, remote control, virtual machines, virtual cameras, browser copilots, display tampering, surfaced by what they do and sealed as evidence, even when the tool is new or renamed.
How detection works →Outpace · off-device
Help that never touches the device
A phone, an earpiece, an accomplice off-camera. No software can see any of it, so the Question Engine changes what they'd be helping with. Included with every account.
How deterrence works →Both run in one place: why interviews need their own client →
Why current detection falls short
Two common approaches.
One shared blind spot.
Most interview-integrity products either infer cheating from behavior or compare running software against a maintained list of known tools. One produces suspicion. The other only recognizes what it already knows.
Capifiq verification layer
Verify the interview environment: not just behavior or a static denylist.
Capifiq is designed to surface unauthorized assistance even when the exact tool is new, renamed, privately built, or adapted from an open-source project.
Behavioral inference
Watches the candidate and guesses.
These tools look for patterns that may correlate with cheating, but the signal remains indirect.
- Eye movement or looking away
- Unusual pauses or response timing
- Head, posture, or attention changes
- Outputs probability or suspicion
!Normal interview behavior can look suspicious, while a careful cheater may look completely normal.
Known-tool process scanning
Matches software against a list.
These scanners inspect known names, signatures, hashes, or window indicators associated with cataloged cheating tools.
- Requires an actively maintained catalog
- Works best on known names and builds
- New tools remain invisible until added
- Renamed or custom variants may evade matching
!A new GitHub project or private build can exist before the scanner’s list is updated.
Behavioral tools infer. Process scanners recognize known indicators. Capifiq verifies and preserves evidence.
Where Capifiq sits
Two different questions.
Two different comparisons.
Products in this market differ on two independent axes: who evaluates the candidate, and where the interview actually runs. Collapsing them into one list is how a buyer ends up comparing an AI interviewer with a meeting platform. Here they are, separately.
Axis one
Who evaluates the candidate.
This is the line that matters most, and it is not the same as "is there AI in the product". Capifiq uses AI for question preparation, session integrity, captions and media authenticity, and none of it forms a view about the person.
| What differs | CapifiqHuman-led interview | AI interviewer platformsAutomated evaluation | Interview intelligence toolsScoring from the recording |
|---|---|---|---|
| Who conducts the interview | A person, liveYour interviewer runs the conversation. There is no autonomous interviewer. |
A modelCommonly asynchronous, with the candidate answering to a recording or an agent. |
A person, liveThe tool joins or ingests the recording afterwards. |
| Who evaluates answer quality | Your interviewerCapifiq does not grade answers. That judgement stays with the hiring team. |
The modelAnswer quality is scored automatically against a rubric. |
The tool, then a personAnswers are summarised or scored, and a reviewer works from that. |
| Fit score, rank or recommendation | None producedNo fit score, no ranking, no advance-or-reject recommendation exists in the product. |
The core outputA score or shortlist is the reason the product is bought. |
Commonly producedScorecards, competency ratings and comparisons across candidates. |
| Automated rejection | Nothing to reject onAn automated screen-out needs a score. Capifiq produces none. |
Configurable in some productsThresholds can gate who a recruiter ever sees. |
Depends on the ATS wiringThe score can drive downstream filtering. |
| Primary output | A session recordTimestamped integrity events and sealed evidence, reviewed by a person who did not run the interview. |
A candidate scoreA rating, band or ranked list. |
A scored transcriptHighlights, ratings and interviewer coaching. |
Axis two
Where the interview runs.
Integrity is a property of the session, so it depends on who owns the session. A tool that watches a meeting it does not own can only see the layer it was given.
| What differs | CapifiqThe meeting is the product | General meeting toolsVideo conferencing | Integrity add-onsBolted onto someone else's meeting |
|---|---|---|---|
| Where the interview runs | In CapifiqScheduled and conducted here. Interviewers join from the browser; the candidate joins through the desktop client. |
In the meeting toolBuilt for meetings in general, not for interviews specifically. |
In the host toolThe add-on observes a session it does not control. |
| Device-session integrity | Verified from the candidate's clientA browser tab cannot see the desktop. The client can, for the length of the interview window only. |
Not providedReasonably so. It is not what a meeting tool is for. |
Bounded by the host appCoverage is limited to what the platform's extension surface exposes. |
| Handling help that never touches the computer | Addressed in the questionsNo software sees a phone or an earpiece. The Question Engine builds questions so that help has far less to offer. |
Out of scopeThe meeting tool has no view of the interview protocol. |
Not addressableA detection layer cannot change what is being asked. |
| Recording, transcript, shared code pad, whiteboard, integrity timeline | One session, one recordThe coding pad and whiteboard are inside the interview, so shared work is part of the same timeline. |
Recording and transcriptTechnical work usually happens in a separate editor or screen share. |
Assembled across toolsThe record is stitched from the meeting platform, the editor and the add-on. |
| Candidate rights and review workflow | Built into the productConsent before capture, adjustment requests, and a reviewer who is barred in code from being the interviewer who ran the session. |
Recording consent onlyThe rest is left to the employer's own process. |
Varies by vendorOften a policy commitment rather than an enforced workflow. |
Capabilities vary by vendor, module and configuration. Several products span more than one column, and some sell an integrity option alongside a scoring one. Compare the specific configuration you are buying, not only the company name. Comparing detection approaches, behavioural inference, known-tool scanning and environment verification, is a third question, covered on Why Capifiq.
AI analysis · live and after
Analyzed live. Re-examined after.
Capifiq's analysis doesn't stop when the interview ends. It works in two passes: real-time during the session, and a deeper AI pass over the complete sealed record afterward, so nothing is missed in the moment and nothing is lost for review.
During the interview
Real-time analysis, live to the interviewer
As the session runs, Capifiq analyzes the interview environment and surfaces events and observations the moment they happen. The interviewer watches integrity status on a live dashboard and can act while the interview is still in progress.
- Live detection of on-device activity
- Behavioral and media observations as they occur
- On-demand evidence capture during the session
After the interview
A deeper AI pass over the full record
When the interview ends, Capifiq runs a thorough AI analysis across the complete sealed record: correlating the full timeline so subtle patterns that weren't obvious in the moment are surfaced in the final integrity report.
- Full-timeline correlation across the session
- Findings compiled into a sealed integrity report
- Evidence retained for later review and audit
Privacy by design
Clear collection.
Explicit controls.
Capifiq documents what can be captured during a monitored interview, which evidence actions are controlled by the interviewer, and how long interview records are retained.
Transparency matters.Evidence collection occurs during the interview session, and collected interview data is retained to match your enterprise retention policy.
View our security and privacy architecture →Every interview
Meeting and screen recordingInterviews run in Capifiq and are recorded, including the interview screens, for the length of the session.Interviewer controlled
On-demand display screenshotsThe interviewer can capture a screenshot of any display at any moment, whether that display is currently shared or not.Always captured
Control and hotkey activityControl-key combinations and monitored hotkey activity are captured throughout the active interview.Enterprise retention
Long-term evidence retentionData collected during the interview, including sealed integrity evidence, is retained to match your enterprise retention policy.Capifiq surfaces the finding. The interviewer makes the judgement. Findings are reviewed by a qualified person and are never the sole basis for a decision.
NIST-aligned evidence handling · SOC 2 Type II audit in progress · audit-ready evidence retained on your terms · candidate rights & review
The complete platform
One platform for the whole interview.
Directly observed technical events at the core, with media-authenticity signals labelled separately: inside the platform the interview actually runs on, connected to your ATS and calendar. One platform for the entire interview, and a record built to be reviewed rather than argued with.
Capifiq runs the interview and the record it produces. It runs the interview itself, so nothing has to be bolted on around it, and delivers deterministic detection of on-device and environment threats, with media-authenticity signals labelled separately from direct technical events; connects to every applicant tracking system and your calendar; adds optional live ID card verification; and runs both live in-interview and post-interview AI analysis across every signal collected: all preserved as one timestamped, cryptographically sealed evidence record. It is not a process-list scanner: rivals advertise a tool count because a scanner only catches names on a list: rename a binary or fork it on GitHub and they go blind. Capifiq doesn't count tools; it verifies the environment, catching unknown, renamed, forked, and private builds alike.
Deterministic detection
- Hidden AI assistants (desktop & web)
- Browser- & API-based AI (no local process)
- Shows what a detected tool is displaying
- Remote-control & screen-share sessions
- Virtual machines, sandboxes & emulators
- Virtual cameras blocked at join: a deepfake feed cannot connect
- Deepfake verification on the live camera feed
- Display tampering & hidden displays
- Live view of every connected display, not just the shared one
- Unauthorized & injected input devices
- Virtualized & isolated environments
- VPN, proxy & location masking
- Device & display anomalies
Analytical signals (media authenticity)
- Audio-visual consistency of the feed
- Deepfake & manipulated video
- Synthetic & cloned voice
- Face-consistency signal on the live feed
Platform & integrations
- The interview runs on Capifiq: nothing bolted on around it
- Shared code pad and whiteboard built in
- Live captions with per-speaker labels
- Panels, observers and email-invited guests
- Every interview recorded and transcribed by default
- ATS and calendar workflow via integrations, API and webhooks
- Optional live ID card verification
- Keep your ATS, calendar, and coding tools
AI analysis, live & post
- Live in-interview AI analysis
- Post-interview AI analysis of every signal
- Full-timeline correlation
- Compiled into a sealed integrity report
Evidence & reporting
- Timestamped event timeline
- Cryptographically sealed reports
- On-demand cross-display capture
- Audit-ready retention
Coverage model
- Direct technical events + labelled analytical signals
- Catches unknown, renamed, forked & private tools: no list
- On-device detection + off-device deterrence
- No score, no risk band, a human decides
Direct technical events are observed and recorded; media-authenticity signals are model-derived and labelled as such: never scores, never about the person. Off-device help is outpaced by the Question Engine. Every gap, covered by one platform.
Simple pricing.
Powerful integrity.
Scoped pilotNo subscriptionCredits never expireUnlimited seats