Glossary

A job application filled in and submitted by software acting for a person, from browser extensions that apply in bulk to general AI assistants that can complete web forms. Many are sent on behalf of real, genuine job seekers.

In Elsin. Per team or per job, employers let agent applications in with a label, hold each until the candidate confirms, or accept verified candidates only.

AI agents applying to jobs

When a tool connected to an applicant tracking system (ATS) sends its result back onto the candidate's record in that ATS, so recruiters see it where they already work instead of in a separate dashboard.

In Elsin. In Greenhouse, Ashby, Lever or Workable, Elsin adds a tag such as Elsin: Verified and a note. A new result replaces the old tag, and no other tag is touched.

ATS integrations

Permission a person gives, before any capture, for their biometric data such as a face geometry to be collected and used for a stated purpose. Laws such as the Illinois Biometric Information Privacy Act (BIPA) require written notice and a written release before collection, and the GDPR treats biometric data used to identify someone as a special category that needs a lawful basis such as explicit consent.

This is general information, not legal advice.

In Elsin. The camera stays off until the candidate has read what happens and agreed. Anyone who would rather not scan their face can send their ID to a person instead.

Any attempt to misrepresent who is applying for or doing a job. It includes invented or stolen identities, someone else sitting the interview, one person applying under many names, and invented work history.

How to spot a fake job candidate

A video interview where the candidate's face or voice is altered or replaced in real time by software. The FBI has warned that North Korean IT workers use face-swapping technology in video job interviews.

What employers can do about deepfake interviews

Checking that an identity document is genuine, readable and consistent: the data printed on it, its machine-readable zone or barcode, its expiry date and its photo. On its own it shows a document exists, not that the person holding it is its owner, which is why it is usually paired with a face match.

Comparing one live face to one reference photo, usually the photo on an identity document, to decide whether they show the same person. It is a one-to-one comparison, which is different from searching a face against a database of many people (one-to-many identification).

In Elsin. The match runs on the candidate's phone. Only the result and a match score are sent. Elsin keeps no face image or template.

An applicant whose identity is invented, borrowed or stolen. The profile, résumé and references can all look normal, because they are often assembled from real people's details or written with AI tools.

Warning signs, and what to do about them

A list of applications held for a person to look at, each with the reason it was flagged. It keeps automated signals in an advisory role, so a false positive costs a few minutes of a reviewer's time instead of a rejected candidate.

In Elsin. Elsin never rejects anyone automatically. A reviewer clears the flag, keeps it, or asks the candidate to verify, and each decision names the person who made it.

Confirming that a person is real and is who they claim to be, usually by checking an identity document and comparing the person to the document's photo. NIST SP 800-63A calls this identity proofing.

In hiring, it answers a narrower question than a background check: is this applicant a real, unique person, and is the same person present at each stage.

In Elsin. Shortlisted candidates check their ID and face on their own phone. The result lasts 12 months and counts at every employer on Elsin.

Identity verification in hiringBackground checks vs identity verification

A fingerprint of a value made with a secret key, such as HMAC-SHA256. The same input and key always give the same output, so two records can be compared, but without the key nobody can recompute the fingerprint from a guessed email or phone number.

In Elsin. For the shared fraud network, emails, phones, devices, résumé fingerprints and document codes are compared as keyed hashes. Only the Elsin server holds the key.

How Elsin stores identifiers

A location, often a private home, where company laptops shipped to remote hires are kept running so that workers abroad can log in remotely and appear to be working from inside the country. US authorities have searched dozens of suspected laptop farms tied to North Korean IT worker schemes.

North Korean IT worker schemes

A test that the face in front of the camera belongs to a live person present at that moment, rather than a photo, a screen replay, a mask or injected video. Active checks ask the person to move, for example to turn their head in a requested order. Passive checks analyze a single capture.

In Elsin. The head-pose check runs in the browser on the candidate's phone. Only timing evidence reaches the server, and no image or video leaves the device.

The two or three lines of capital letters, numbers and chevrons (<) at the bottom of a passport photo page or on the back of many ID cards. The format is set by ICAO Doc 9303 and carries the holder's name, document number, nationality, date of birth, expiry date and check digits, so software can read and validate it.

Many US driver's licences carry the same kind of data in a PDF417 barcode instead.

In Elsin. The phone reads the document number from the MRZ or barcode, or the candidate types it, and hashes it on the device. The raw number is never sent.

Risk indicators read from how an application arrives, without asking the applicant to do anything: the network it came from (for example a VPN or a hosting provider), a disposable email address, a form completed faster than a person could type, or a device, phone number or near-identical résumé shared with other applicants.

A passive signal is a reason to look closer, never proof of fraud on its own.

In Elsin. Signals add to a risk score with the reason written out. The most a score can do is send someone to a person for review.

What the risk score is made of

An interview where someone other than the applicant answers the questions, either on camera in their place or by feeding answers off screen. The person who later starts the job may not be the person who interviewed.

In Elsin. A quick re-check before the interview and on day one asks the candidate to scan the same ID and match a fresh face to it on their phone.

Deepfake and proxy interviews

Data pooled across several organizations so that an identifier flagged at one can be recognised at another. The privacy risk is that it can expose one employer's candidates to another, so the design matters: what is shared, in what form, and what a match reveals.

In Elsin. A match at another employer returns a count only. No names, records or employer names are shared.

How the shared fraud network works

An IETF effort to let automated clients prove who they are by signing their HTTP requests with HTTP Message Signatures (RFC 9421) and publishing their public keys. A website can verify the signature to know which agent sent a request. It identifies the agent, not the person the agent works for.

In Elsin. Elsin checks these signatures, alongside known agent user-agent tokens, to label an application as sent by an agent. The label never adds to a risk score.