How does Codility protect assessment integrity?

Identity verification, proctoring, behavioral signals, hidden app detection and a similarity check produce integrity signals for human review, alongside the submitted work and a deterministic risk rating. Available controls depend on the assessment type and configuration. No single signal proves misconduct, and no candidate is rejected automatically. See Screen for engineering

Can candidates use ChatGPT or another outside AI tool?

The hiring team sets the rules. Enabling Codility's AI Assistant gives candidates a provided assistant with every interaction saved. When AI is off, behavioral signals, hidden app detection and similarity checks can flag possible outside tool use. A person reviews the evidence.

Can candidates use Claude Code or Codex in an assessment?

Yes, when you allow it. Claude Code from Anthropic and the Codex CLI from OpenAI run inside the Codility assessment environment in Screen, Interview and Skills Intelligence. They follow the same AI Assistant settings: on or off for each task in Screen and Skills Intelligence, and for each interview in Interview. The full interaction is saved to the candidate record for review.

Is using AI during an assessment cheating?

The assessment rules define permitted AI use and are shown before the candidate starts. When Codility's AI Assistant is enabled, its interactions are recorded for review. If outside tool use is flagged, a reviewer assesses the evidence against those rules.

Do you record candidates?

Only when the hiring team enables proctoring for that assessment, and candidates are told before they begin. Screen captures snapshots or a recording. Interview records the session with a full transcript. Proctoring recordings are retained for 30 days. See security and compliance

What happens when a signal is flagged?

The report shows the signal and its contribution to the risk rating. A reviewer checks the replay, snapshot, similarity match or AI usage, then decides whether to follow up. No action against the candidate is automatic.

Does integrity monitoring work the same for business tasks?

The same report, risk rating and human review apply. The evidence differs with the work: a business record holds the submitted response and every prompt, and code replay belongs to engineering tasks. Available controls depend on the assessment type and configuration.

When you suspect cheating, see what actually happened

See integrity signals, the submitted work and every prompt in one report. People review the evidence and make every decision.

Request a free trial

Take the product tour

Validated assessments since 2009 SOC 2 Type II ISO 27001

Integrity risk ratingModerate2 signalsSample report

Code replaytask 2 · 41 min

0:0041:00

def parse_rows(lines):

return [r.split(“,”) for r in lines if r]

orders = parse_rows(raw)

Same helper the candidate wrote in task 1, line 14

Similarity check: no matchIdentity verified

A risk rating is never proof. A person reviews the evidence and decides.

Reviewer decision10:41

The paste is the candidate’s own code from task 1. Advance to interview.

A. Moreno · reason recorded

Decide whether AI is allowed before anyone starts

In Screen and Skills Intelligence, switch AI assistance on or off for each task, so one assessment can test skills with and without AI. In Interview, set it for each session. Every prompt and response is saved for review. Account administrators can also lock the setting or switch off every AI feature across the account. Whatever the setting, you can show a reviewer which posture applied.

Claude Code and Codex run inside the assessment, under the same AI settings, and every interaction is saved for review.

When AI is allowed, review prompts, responses and how the candidate used them alongside the submitted code. When AI is off, integrity signals can indicate possible outside tool use for a person to review.

See how Interview records a live session

Assessment report showing the candidate's submitted code on the left and the saved AI Assistant and Claude Code conversation on the right, sample data.

Screen report summary with an assessment integrity card listing identity verified, proctoring on, a similarity match and an AI overlay detected, sample data.

Code replay timeline for one task, with markers for paste events, time away from the tab and test runs, sample data.

The signals reviewers see

Each control is designed to give a reviewer a signal beside the work, from identity checks to code replay and AI transcripts. Available controls depend on the assessment type and configuration.

Identity verification. Photo ID matched to the person taking the assessment
Proctoring. Snapshots or a recording in Screen, a recording and transcript in Interview
Behavioral signals. Browser absence, tab switching, paste volume and typing patterns
Hidden app detection. Where enabled, the desktop app is designed to surface tools hidden over the screen
Similarity check. Compared with other submissions and solutions found online
Code replay. The solution being written, with markers on the timeline
AI Follow-Up Questions. Short questions on the work, kept out of the score
AI usage. Every AI Assistant, Claude Code or Codex interaction, with the AI skills used

A risk rating is never proof

Assessment reports show an integrity risk rating: None, Low, Moderate or High. Fixed rules combine the signals, so the same signals produce the same rating.

Review the contributing signals alongside the work before deciding. Codility never rejects candidates automatically. No single signal proves misconduct.

AI Follow-Up Questions are deliberately kept out of the score, so the answers inform the reviewer and never change the result.