What a Proctored Assessment Reaches on Your Machine

What Can It See, Tier by Tier?

Vendors publish the answer on their own product pages, and it runs from a browser watching itself to a full recording of you and your screen. HackerRank's Screen page lists tab proctoring, copy and paste tracking, multiple-monitor detection, leaked question alerts and full session replay ¹. It adds an AI proctor that intervenes only when suspicious activity is detected, and an image analysis system that verifies identity and monitors for multiple faces during an assessment ¹.

CodeSignal's page describes the heavier end. Its proctoring records video, audio and screen activity for the entire assessment session ².

The identity step prompts you to present a valid government-issued photo ID, matched to the person on camera, and passes the verification status to the hiring team with your results ². The same page describes a suspicion score built from solution similarity, telemetry and copy-paste activity.

Which of those you meet is the employer's choice.

Both pages describe the assessment session itself, and neither says which settings are defaults. Neither publishes an accuracy figure or a false-positive rate for any signal it names. What either system reaches when the session is over is not on either page.

A product page advertises what a product can do, and it is not an audit of how well it does it.

Read the Consent Screen Before You Accept

The consent screen is where an assessment can tell you which of those settings it turned on; if it does not, ask the employer. HackerRank's candidate AI notice says its AI features may analyse coding, behavioural signals, webcam images and screen activity to detect anomalies, impersonation and unauthorised use during assessments and interviews ³. Proceeding with an AI feature is taken as acknowledgement of that notice ³. The switches are the employer's; the notice is the platform's.

Declining is a real option with a real consequence, and the consequence differs by vendor.

HackerRank's knowledge base says that when a customer enables an automated tool it secures the candidate's consent first, and that a candidate who does not give consent will not be able to participate in the feature using that tool ⁴. The same page says that under New York City's Local Law 144 it is the employer that must offer the candidate the alternative-method request, and says nothing about what happens elsewhere.

That page sends the question of an alternative to the employer.

CodeSignal's page does not say how long a recording is kept. HireVue, the one vendor here that answers it in general terms, leaves it to the employer. Its candidate FAQ says it stores personal data for a retention period chosen by the company you interviewed with, purges it automatically when that period expires, and that you can request deletion at any point in the process ⁵. The vendor states plainly that it does not decide whether or when to delete; the company you applied to does.

What an employer is told to disclose before recording a screen runs to six items: what is captured, what is not, why, who watches, how long it is kept, and what declining gets the candidate.

If the invitation you were sent covers none of those, asking is cheap.

What Happens When a Flag Goes Up?

A flag goes to a human, and the vendors say so themselves. HackerRank's knowledge base on AI plagiarism detection says its model analyses code-writing patterns, time taken, copy-paste activity and tab-switching behaviour, and classifies suspicious attempts as High or Medium ⁶. Flagged attempts are highlighted in candidate reports, and recruiters can replay keystrokes to see why a candidate was flagged ⁶. The feature is stated to cover coding questions only ⁶.

The vendor puts a number on how often a flag is wrong.

The page states the model's overall precision is 85%, meaning flagged sessions are correct 85% of the time, and that human oversight therefore remains crucial ⁶. By that figure, roughly one flagged session in seven is a mistake.

There is no test set, no period and no base rate on that page, and no recall figure, so it says nothing about how many real cases go unflagged. CodeSignal describes the same shape from the other side: sessions are evaluated with automated analysis, with human reviewers validating flagged activity and final decisions, and results returned within one hour ².

A High or a Medium is a claim about your session that somebody still has to read.

Neither page publishes an appeal route for a flag. What the hiring team does with a High or Medium is the hiring team's decision, and it is not on either page.

What Can You Ask For Afterwards?

Four requests are named in HackerRank's candidate AI notice, and each one is addressed to the employer. It says you may request an alternative selection process or a reasonable accommodation, and human review of a decision made or substantially assisted by AI ³. You may also request correction of inaccurate personal information, and an explanation of how AI was used if an adverse decision is made about you ³.

Requests are all the notice creates, and it does not say an employer has to agree.

Where no law creates the right, employer policy is all that stands behind it. The addressee named in the notice is the employer rather than the platform, so the message goes to the recruiter or the contact named on the invitation.

The same notice treats biometric information separately. It says any biometric information requires separate informed written consent, and is permanently deleted when the purpose for collection is satisfied, within three years of the last interaction, or as required by law, whichever comes first ³.

The purpose is typically set by the employer, so the real period is the employer's and the notice does not state it.

Where a disability is behind it, a request for a different format can also be an accommodation request, and the notice names that route in its first item. Asking in writing, to the contact on the invitation, is the whole mechanism. None of it happens automatically.

Who Pays for the Surveillance?

The person sitting it does, and part of that cost has been measured in students sitting proctored exams. Four waves of survey data covered 1,760 students in 105 courses at one faculty of a Dutch university ⁷. Proctoring had no measurable effect there on self-reported temptation to cheat, on perceived exam difficulty, or on performance, and students reported higher test anxiety under it ⁷. Courses chose whether to proctor, so this is observational data rather than a trial ⁷.

The same study found the anxiety effect was worse for students without a dedicated study space, without reliable technology, or under financial stress ⁷.

Sixteen interviews with university students who hold documented disability accommodations and have sat online proctored exams describe anxiety about how surveillance interacts with their disability, fear of being misread as cheating, and added cognitive load during the exam ⁸. Students in that study describe compromises they make between privacy and accommodations in order to take a remote test ⁸.

Both are education, not hiring, and neither can be read as a measurement of a hiring process. Temptation was self-reported rather than behaviour observed, so the Dutch study is not evidence that proctoring fails to reduce actual cheating. Sixteen interviews establish that these harms occur and describe their shape; they cannot say how common they are.

An accommodation request should not be the price of not being flagged.

Employers meet this as a design question rather than a settings question, and one page here sets out what they are told: it weighs proctoring against an assessment that permits AI. Another, written for you, covers whether you have to agree to be recorded, which is the question sitting in front of you.

See what gets scored >