The Sherlock AI alternative
Sherlock listens to the room.
A phone read in silence makes no sound.
Sherlock classifies device, audio and behaviour signals and alerts your interviewer mid-interview. Capifiq stops trying to hear the help and makes it useless instead.
Side by side
Capifiq and Sherlock AI, side by side.
| What matters | CapifiqInterview integrity | Sherlock AIA classifier beside your meeting tool |
|---|---|---|
| Help from a phone, earpiece or accomplice | Made uselessQuestions a snapshot of the screen cannot answer. | Listens to the roomA phone read silently makes no sound. |
| Cheating software on the candidate's computer | Caught by what it doesBrand-new, renamed and GitHub-copied tools too, and remote control. | AI copilots claimedRemote control of the computer not listed. |
| ID and face verification | Optional, before the interviewID matched to the candidate's face; your interviewer confirms the same person joined. | ClaimedDescribes 'continuous identity verification'. |
| Deepfakes | Checked on the live feedVideo and voice checked for fakes. Never a score. | ClaimedFlags deepfakes in real time. |
| Where the interview runs | In CapifiqVideo, code pad and whiteboard in one session. | Your meeting toolA toggle on Zoom, Meet or Teams. |
| When your interviewer finds out | After the interviewNothing steers the conversation midway. | A mid-interview alertThe rest of the interview is held under suspicion. |
| What you receive | A timestamped recordNo score, no ranking, no verdict. | An automated classificationOne machine judgement about the candidate. |
| Reading the candidate's behaviour | NeverNo AI judges the candidate. Only deepfake checks analyse their video and audio. | One of three inputsReads 'candidate behavior' alongside audio and device activity. |
| Keeping your current meeting tool | Capifiq's own roomCandidates join the Capifiq session. | YesRuns on Zoom, Meet or Teams. |
Based on Sherlock AI's own published material as of 26 October 2026. Compared on capability, not price.
Where cheating happens now
The help that matters most never touches the computer.
A phone below the webcam, an earpiece, someone out of frame. No software on the interview computer can see them, ours included.
Sherlock's answer: listen to the room.
It combines "signals from device activity, audio environments, and candidate behavior" into one classifier. Help spoken aloud leaves something to hear. A phone read silently, or an earpiece, leaves nothing.
Capifiq's answer: make the phone useless.
The Question Engine builds the questions. A snapshot of the screen is not enough to answer, and the follow-up your interviewer asks was never on screen. Relayed help becomes slow and visible in the conversation.
How Sherlock AI detects
A classifier that alerts while you interview.
Sherlock describes a "multimodal adversarial ML approach" feeding a "unified classifier", with "real-time alerts" to the interviewer.
Silence beats a microphone
A phone read silently and an earpiece make no sound in the room. A classifier built on audio has nothing to classify.
Behaviour is the weak input
'Candidate behavior' varies with disability, nerves and the room someone sits in. Feeding it to a classifier flags those candidates more often.
An alert cannot be unseen
Once your interviewer is told to be suspicious, every pause reads as evasion. Whatever follows is no longer independent evidence.
The number on their homepage
What "~85% to over 97%" means in practice.
Take the top of the range: right 97% of the time in both directions. Run a thousand candidates through it.
| Candidates who actually cheat | Cheats flagged | Honest people flagged | Flagged but innocent |
|---|---|---|---|
| 10 in 100 cheat | 97 | 27 | about 1 in 5 |
| 5 in 100 cheat | 48 | 29 | more than 1 in 3 |
| 2 in 100 cheat | 19 | 29 | 3 in 5 |
| 1 in 100 cheat | 10 | 30 | about 3 in 4 |
At two cheats in a hundred, three in five flagged candidates are innocent, and each was flagged mid-interview.
The legal side
HireVue agreed to pay $3.75 million. Ask who carries the risk.
In 2026 HireVue agreed to a $3.75 million settlement of an Illinois class action alleging its video interviews collected candidates' voice and facial biometrics without the written consent the state requires. HireVue denies wrongdoing. Sherlock's classifier draws on the interview's audio and on candidate behaviour.
Biometric consent
Illinois law requires written consent before faces or voices are analysed, and lets candidates sue for $1,000 per negligent violation and $5,000 per intentional one. Any tool that reads faces or voices raises the same question: who obtained that consent?
Emotion recognition
Since February 2025 the EU AI Act has banned AI that infers emotions at work, recruitment included, with fines up to 35 million euros or 7% of global turnover. The ban applies to the employer. Ask any vendor that reads voice stress or expressions whether it is caught.
Automated decisions
Tools that score or screen candidates bring bias audits and candidate notice (New York City Local Law 144). In 2023 iTutorGroup paid $365,000 to settle an EEOC case after its software rejected older applicants. The employer paid.
How Capifiq handles it
Capifiq treats every interview recording as biometric, asks candidates for a separate written release before any interview involving covered biometric processing, and publishes its retention schedule.
Settlement details as of 14 November 2026. This is not legal advice; obligations depend on jurisdiction and use.
Why teams switch
Why teams choose Capifiq over Sherlock AI.
It writes the interview for you
The Question Engine builds the interview from your job description, and you approve every question. A screenshot is not enough to answer, and the follow-up was never on screen.
Catches tools no list has heard of
Not a list of known cheating apps. A patent-pending, physics-based engine catches tools by what they do, so brand-new, renamed and GitHub-copied tools are caught too.
Evidence, not a verdict
A timestamped record your interviewer reads. No score, ranking or risk profile, and nothing for a regulator to classify.
Questions
Common questions.
Can Sherlock AI detect a phone or earpiece?
Why is a real-time cheating alert a problem?
Why is Capifiq a lower-risk choice than AI interviewers and behavioural proctoring?
What changes for our interviewers and candidates?
Does Capifiq use behavioural analysis or a room camera?
Which laws apply when a tool scores candidates?
See it on your own interviews
Ready to see the difference?
Bring a real role. We will build the interview with the Question Engine, run it in Capifiq, and show you the record you would have at the end.