Is the Amazon Online Assessment Proctored?
By Aaron Cao · Updated

Amazon does not publish the monitoring setup for its Online Assessment, and it varies by role, region, and year. Most software candidates report no live human proctor and no camera requirement, but the invitation email states the rules for your specific assessment. Read it before assuming.
What does proctored actually mean here?
The question usually comes with an unstated one behind it, and the honest version of this page has to answer both. This section separates the two, because the first has a partial answer and the second has a clear one. Proctoring is a spectrum, not a switch, and candidates asking "is it proctored" often mean four different things at once.
At one end is live human proctoring, where a person watches your camera feed in real time and can stop the session. In the middle sits recorded proctoring, where the camera and screen are captured for later review, and automated flagging, where software scores your session for anomalies without a human present. At the far end is plain telemetry: the assessment platform logging what happens inside its own page, which every online assessment does by definition because it is a web application recording your work.
So an assessment can have no proctor in the everyday sense and still produce a detailed record of how you worked. Confusing those two is where most bad decisions get made.
What does Amazon say, and what does it not say?
Amazon does not publish the monitoring configuration of its Online Assessment, and there is no public specification to cite. What is consistent is that the assessment arrives by email with instructions, a time limit, and a code of conduct you accept before starting. That email, plus the assessment platform's own instructions page, is the authoritative source for your assessment. A thread describing someone's OA from a different year, a different country, or a different role is not.
Candidate reports for the software engineering assessment commonly describe a browser-based, timed test taken alone, without a live proctor and without a webcam requirement. Those reports are consistent enough to be useful and weak enough that you should not build a plan on them. Amazon runs assessments for many role families and has changed vendors and formats over time; an assessment for a non-engineering role or a different region may look nothing like the one your friend took.
The reliable move is to read the invitation carefully and, if anything is ambiguous, ask your recruiter. Asking is normal and carries no penalty. Other pages on what specific assessment platforms do sit in the detectability library.
What can the platform see without a camera?
More than the absence of a webcam suggests. An assessment runs inside a web page the company controls, so it sees what that page can observe: when the session starts and ends, how long each problem took, how the code evolved between submissions, when the tab lost and regained focus, and what your submitted solution looks like next to everyone else's.
The last one matters most and is the least discussed. Submissions can be compared against other candidates and against public solutions, and the comparison does not require a camera or a proctor. It is a query run later, quietly, on data the platform already has.
Paste behavior sits in the same category. Whether any particular platform records it is not something to assume either way, but it is trivially observable inside a browser page, and it costs a company nothing to log. Treating anything that happens inside the assessment page as observable is the only assumption that does not age badly.
Why the OA comes back in the interview
Here is the part that makes the proctoring question less important than it seems. Passing the assessment is not the last time your code exists. Interviewers frequently have your OA solution in front of them in a later round, and asking you to walk through a design decision in it is a normal, easy question for them to ask.
A new graduate interviewing for an entry-level software role cleared an assessment with a solution he had not fully reasoned through. In the follow-up round an interviewer asked why he had chosen a hash map over a sorted structure in one function. He could not say, and the interview did not recover, because the gap between the written work and the spoken explanation is the single loudest signal in a technical loop. Nothing needed to be detected; the conversation did the work.
This is also where an honest statement about tools belongs. A live interview assistant listens to an interviewer speaking and suggests structure while you talk. An online assessment has no interviewer, is often explicitly covered by a code of conduct you accepted, and is exactly the context where no assistant is in scope, alongside proctored tests, recorded screen shares, and company-managed devices. What genuinely transfers is rehearsing the spoken explanation of your own solutions, which is what the mock interview page is for. Our own data handling is documented on the security page.