CodeSignal does not issue a cheating verdict. It runs two separate checks and hands the results to people. The first is a human review of the proctoring recording, which can reject a session for a documented list of reasons. The second is an automated Suspicion Score that rates each question submission from "None" to "High" based on paste events, description copies, code similarity, and timing. As of September 9, 2026, CodeSignal's own guidance to employers says "you shouldn't treat a high Suspicion Score like a guarantee that a test-taker was cheating." This article covers what each check looks at and what happens when one fires.
Check one: the proctoring review
On a proctored assessment, CodeSignal records camera, microphone, and screen for the whole session and a verification team reviews it afterward. The recruiter guide says "CodeSignal verifies the result within 1-3 business days" and lists the common reasons a result is not verified:
- "Test-taker was not alone in the room during the assessment"
- "Test-taker got up from their seat during the assessment and exited the camera's view"
- "Test-taker used other devices during the assessment"
- "Test-taker used an external IDE"
- "Test-taker searched online for help (other than syntax-related searches)"
Each of those is a rule you acknowledge at setup. The outcome is binary: the result is verified, or it shows the recruiter as "Proctoring Rejected" with a reason on hover. The score itself is unchanged; CodeSignal's proctoring documentation says "Proctoring has no impact on the score you receive." What the recording contains is inventoried in what CodeSignal proctoring records.
Check two: Suspicion Score
This runs on every submission, proctored or not, because it uses editor data rather than video. CodeSignal's Suspicion Score documentation describes it as flagging "potentially suspicious activity from a variety of sources" to "assign a trust level for each question submission."
The recruiter sees a top-level "Integrity Flagged" value of Yes, No, N/A, or Pending, and then a per-question score. CodeSignal's guidance on what "Yes" means: "'Yes' does not mean a test-taker definitely cheated, but it means there's enough suspicious activity that you should consider looking into that submission more closely."
The signals CodeSignal names, with the patterns it tells reviewers to look for:
| Signal | What CodeSignal says raises suspicion |
|---|---|
| Similarity Score | Solution matches other submissions; reviewer uses "View Diff" to judge whether the problem is simply small |
| Suspicious Paste Events | "Larger chunks and/or more frequent pastes are more suspicious" |
| Paste timing | "Pasting after having no activity in the replay for a while, and/or after a Description Copy event is a common cheating pattern" |
| Paste plus failed submits | "Multiple repeat pastes followed by submissions resulting in errors is a common cheating pattern" |
| Language switch | "Switching the language combined with paste events is also a highly suspicious behavior pattern" |
| Description Copy | Copying the prompt, especially early, then typing later "can be indicative of retyping a solution" |
CodeSignal says these flags "could mean the test-taker was copying off another test-taker's answers, looking at leaked materials, or using tools like generative AI to come up with code." The reviewer sees them in an editor replay that marks paste events and skips idle time, per the results documentation.
Note what is absent: tab switching is not a named Suspicion Score input, and there is no documented keystroke-cadence model. The paste and copy mechanics are covered in whether CodeSignal detects tab switching and copy-paste.
What a flag proves
Less than it looks. CodeSignal tells employers directly: "Innocent behaviors can sometimes look suspicious, and different companies have different perspectives on the use of outside tools." Pasting your own boilerplate after reading the prompt, or switching language after a false start, produces the same markers as copying a solution.
Two things are true at once. A flag is not a rejection, and you will not be told about it. The Coding Report goes to the requesting company, which decides what to do. On a proctored test, a rejected review means the result is not certified, and the cooldown rules then govern whether you can attempt again; see CodeSignal cooldown and retake rules. How to respond if a recruiter asks you to explain a solution is covered in what happens when a coding assessment flags you.
Can you use AI on CodeSignal?
Not on a certified assessment. The GCA rules article allows syntax-only web searches and then states: "This also means that the use of AI is not allowed, including syntax search." The acknowledgement reads "I will not use external generative AI tools." A company custom assessment or interview can set different rules; the acknowledgements on your screen are the authority, and when AI is allowed in a technical interview covers how to confirm in writing.
What this means for candidates
Write in CodeSignal's editor from the first line rather than drafting elsewhere and pasting; a paste is the most-cited signal in CodeSignal's own guidance. Stay in one language unless you have a reason you could explain. On a proctored test, follow the surroundings rules literally, since the review is a person watching for exactly those five reasons. If the rules prohibit AI or outside resources, prepare with them beforehand and take the test without them.
Share
Explore with AI
Continue exploring
Related guides
Assessment Guides · 5 min read
Does HackerEarth Detect Cheating? What It Flags
HackerEarth flags code matching another candidate above 85%, runs AI checks against public sites, and scores an integrity index. What each flag proves.
Assessment Guides · 5 min read
Does Mettl Detect Cheating? Flags, Proctors, and the Index
Mercer Mettl rolls browser rules, eight AI flags, optional live proctors, and a code similarity check into a credibility index. What each catches, what a flag means.
Assessment Guides · 5 min read
Does HackerRank Detect Cheating? What It Flags in 2026
HackerRank flags sessions with an 85%-precision plagiarism model, tab and paste tracking, and screenshot analysis. What each layer catches and what a flag means.