No single flag proves a candidate cheated. Combine the signals, watch the replay, and ask the candidate to explain their own code before you decide.
The short answer: combine signals, confirm live
Every vendor says the same thing in its own words. CodeSignal says a "Yes" on its Integrity Flagged field does not mean a test-taker definitely cheated. Codility says a flag is not a guarantee of cheating. CoderPad says unusual activity is not necessarily proof. Treat any one flag as a prompt to look, not a finding.
Signals worth checking
- Similarity to other submissions or public code
- Paste events, especially a large paste after a stretch of no activity
- Tab or IDE exits during the test
- The session replay
- A sudden drop in solve time on a specific question
- Identity and IP changes
None of these stands alone. Several together raise the odds.
What the platforms actually report
CodeSignal's Suspicion Score rolls flags into Integrity Flagged (Yes, No, N/A, Pending) and per-question scores from None to High. It warns that innocent behaviors can look suspicious.
HackerRank runs MOSS similarity on all tests by default, comparing code structure, so renaming variables is generally ineffective. It grades matches High (90%+), Medium (80 to 90%) and Low (under 80%), and the recruiter keeps or removes each flag. Its optional AI model cannot determine the exact source of plagiarism.
Codility reports photo ID on premium plans, network IP detection, a similarity check against past submissions and leaks, and optional proctoring.
Coderbyte summarizes plagiarism as Not detected, Likely or Detected, and adds video playback and a Google search log.
CoderPad Screen reports playback of exits and pastes, similarity alerts, IP tracking, IDE exit tracking and optional webcam proctoring. Does Glider AI record your screen covers how another platform handles the same question.
Where false positives come from
A high similarity score on a simple question can mean any reasonable solution looks alike. Switching from home Wi-Fi to mobile data trips an IP check. VPNs and travel explain location changes.
AI-generated code is the harder problem. HackerRank's own blog showed a ChatGPT solution passing MOSS cleanly, and said MOSS has high false positive rates. HackerRank's 93% accuracy figure for its AI model is a vendor claim, not an independent benchmark.
The fair follow-up: a live walkthrough
CoderPad suggests discussing the playback in a technical interview. A candidate who cannot explain the code likely did not write it. Coderbyte offers webcam video explanations.
Ask about a design choice, a trade-off, and what they would change. Walk through one function line by line. If the candidate has already taken the test and failed it, what happens if you fail a pre-employment assessment explains what they experience. Before you treat a flag as a fail, can you still get hired if you fail an assessment test shows what the result means on their side.
Write your AI and resources policy before the test
CodeSignal's score may flag behavior your company allows. CoderPad's sample candidate message permits ChatGPT but requires working alone. Decide the rule first, put it in the test instructions, and score every candidate against the same standard with AI Candidate Evaluation.
Then review flags against that published rule.