Screen Better. Hire Smarter.

Watch the machine. Never the person.

AI interview-assist tools run as invisible overlays — outside the shared screen, listening to your questions and feeding the answers. Talon shows you what's actually running on the candidate's machine — eliminating all suspicion and doubt.

No screenshots · no keystrokes Zero candidate admin rights Windows agent today No screenshots · no keystrokes Zero candidate admin rights Windows agent today No screenshots · no keystrokes No screenshots · no keystrokes Zero candidate admin rights Windows agent today No screenshots · no keystrokes Zero candidate admin rights Windows agent today No screenshots · no keystrokes

The interview looks fine.
It's by design.
And that's the problem.

SESSION ACTIVE 4 SENSORS
Overlay rendering 001

The AI tool draws over the screen.

Renders on top, outside the share region. You are looking at a window that can't be detected by screen sharing.

Audio routing 002

It hears the question.

A virtual cable routes your audio to a listener. It answers before the candidate does.

Remote desktop 003

It isn't their machine.

The desktop you're watching is a live stream. The real work happens somewhere you can't see.

VM environments 004

The interview runs in a clean room.

VirtualBox, VMware, QEMU — a disposable environment built to leave no trace on the host.

91% of recruiters have spotted candidate deception Greenhouse, 2025 AI in Hiring Report
59% of hiring managers suspected a candidate of using AI to misrepresent themselves Checkr, 2025 survey of 3,000 hiring managers
19% are confident their process would catch a fraudulent applicant Checkr, 2025 survey of 3,000 hiring managers
29% of candidates used AI to answer assessment questions Gartner, 4Q24 survey of 3,290 candidates

Talon — Interview Integrity Platform for the AI era.

Talon by ScoutHawk is an Interview Integrity Platform that helps recruiters make evidence-based, informed decisions for remote hiring. Unlike traditional proctoring solutions, it integrates with existing hiring workflows — giving recruiters the confidence to make the right hiring decisions faster.

Renaming the file doesn't work.

A candidate may rename a tool to Chrome.exe, but the binary still declares its real name. Talon compares the two and escalates discrepancies.

CANDIDATE MACHINE ON-DISK NAME chrome.exe TALON ✓ NAME COMPARED PE HDR SIGNER SEVERITY REAL NAME cluely.exe · HIGH RENAMED UNSIGNED
WHAT THE INTERVIEWER SEES
a3f8c2…  unattributed  ——
cluely.exe  Cluely · High
b71e09…  unattributed  ——
WHAT THE CANDIDATE SEES
a3f8c2… → 1password.exe  not reported
cluely.exe → Cluely  reported · High
b71e09… → spotify.exe  not reported
SHA-256  ·  NAMES NEVER CROSS

You get the complete picture / not the candidate

Talon hashes every process associated with the interview session and flags AI-assist tools. All other processes are encrypted to ensure complete data privacy for candidates.

PROCESS

Match the binary.

Known assistant tools by exact executable match — Cluely, Interview Coder, Final Round AI, LockedIn AI, etc.

OVERLAY

See the layer.

Topmost, layered windows that render outside the region you're sharing.

AUDIO

Follow the cable.

Virtual cables — VB-Audio, Voicemeeter — that pipe your question to a listener.

ENVIRONMENT

Read the room.

Virtual machines and remote desktop: VirtualBox, VMware, QEMU, TeamViewer, AnyDesk.

Talon works in four steps.

All it takes is 1—2 minutes at the beginning of the interview session.

Severity Weights
Severity weight → score contribution

High           ████████████████████ 32
Medium         ███████████░░░░░░░░░ 18
Low            ███░░░░░░░░░░░░░░░░░ 5

Severity drives the score. Same band logic
runs the live view, the list, and the report.
Composite score

composite   = min(100, Σ delta)

bands
0 Clear  ·  <25 Low  ·  25–59 Medium  ·  ≥60 High

21 tools detected, versioned server-side.
Signature set v2026.07.05
Detection latency

Detection   ██████░░░░░░░░░░░░░░ ~20 s

end-to-end, batch transport

The agent batches observations and posts
every ~10 s; the dashboard polls on the
same cadence.
Evidence report

Observation process "cluely.exe" present
Attribution Cluely · High · exact match
Contribution +32 of 100

Observations and attributions stay
separated. Scoring method ships in the
report appendix.

Change is the only constant — it's never a one-and-done job.

New AI interview assistants will keep emerging. Talon stores raw observations and interview-integrity breach flags, so a new tool detected today re-scores every session we have ever run — and regenerates the report.

[ 01 ]

Stores raw observations.

Interview-integrity breaches are derived in real time, never baked into the record. New AI interview assistants are flagged the moment they are detected.

[ 02 ]

Updates without a download.

Detection runs server-side; the agent ships raw observations and holds no detection logic. New AI-assistant signatures go live without you or the candidate downloading or updating anything.

[ 03 ]

Regenerates the evidence log.

Reports rebuild from stored signals for the full retention window. What you hand a lawyer in month three is derived from what the machine saw in month one.

We provide the evidence. You decide.

Exact matches
Score full weight
Heuristic matches
Confidence-downgraded
Allowlist
Browsers, editors, Zoom suppressed
Methodology
Ships in the report appendix
Observations
Separated from attributions
No verdicts
You make the call
talon://session/4471 · live
λ talon report --session 4471
  ████████╗  █████╗  ██╗      ██████╗  ███╗   ██╗
  ╚══██╔══╝ ██╔══██╗ ██║     ██╔═══██╗ ████╗  ██║
     ██║    ███████║ ██║     ██║   ██║ ██╔██╗ ██║
     ██║    ██╔══██║ ██║     ██║   ██║ ██║╚██╗██║
     ██║    ██║  ██║ ███████╗╚██████╔╝ ██║ ╚████║
     ╚═╝    ╚═╝  ╚═╝ ╚══════╝ ╚═════╝  ╚═╝  ╚═══╝
Observation   process "cluely.exe" present       14:22:07
Attribution   Cluely · High · exact match · conf 1.0
Contribution  +32 of 100 · ranked #1 factor
Report ready. Screenshots: OFF · Keystrokes: OFF · Audio: OFF
λ talon export --json
Evidence Log
v2026.07.05 21 Tools
Exact matches
Score full weight
Methodology
Ships in the report appendix
Heuristic matches
Confidence-downgraded
Observations
Separated from attributions
Allowlist
Browsers, editors, Zoom suppressed
No verdicts
Humans make the call

Start your free trial now.

5 interview credits when you sign up. One credit = one monitored interview. Credits exhaust in 14 days.

Team accounts share an org credit pool — invite your interviewers, keep sessions visible across the team, isolated from every other org.

Book a Demo Team and volume pricing — talk to us. We're onboarding early adopters now.
FAQ

Frequently asked questions.

01Does the candidate need admin rights?

No. The agent runs unprivileged — no installer, no restart.

02Will Windows warn the candidate on download?

Yes, currently. SmartScreen flags our executable because it isn't code-signed yet — that's in progress. The invite email includes first-run guidance. We'd rather tell you now than surprise your candidate.

03Does it work on macOS or Linux?

Not yet. The Windows agent ships today; the overlay and audio sensors use Win32 APIs. macOS is on the roadmap.

04Can a candidate defeat it by renaming the executable?

No. A renamed binary still declares its original name internally, and we compare the two. If the file is also unsigned, we escalate the severity.

05What if they just use their phone or a second laptop?

We won't catch that, and no on-machine agent can. We detect on-machine assistance — where the tooling, the speed, and the convenience are. Off-machine cheating is slow and manual enough that a good interviewer usually notices.

06Does it record the interview?

No. No video, no audio, no screen, no keystrokes. Process names, window attributes, audio device names, VM markers. That's the whole list.

07What happens to the data?

Encrypted in transit, retained 90 days by default (you configure it), auto-purged after. Reports regenerate from stored signals until the window expires.