The Control Journal
Assessment GuidesSeptember 3, 20267 min read

What Codility Detects in 2026: Proctoring and Typing Patterns

Codility's integrity stack now adds typing-pattern flags and a Preview device check. What each signal records, what it misses, and what a flag means.

CControl Editorial Team · Updated September 10, 2026

Codility can record considerably more than the code you submit, but almost none of it is automatic. Proctoring, typing-pattern analysis, a device check, and plagiarism comparison are separate features that the hiring company switches on per assessment, and the combination you face is set before your invitation is sent. This article describes what each signal captures according to Codility's own candidate and recruiter documentation, checked on 3 September 2026, and what a flag actually does once it appears on your report.

The single most important change since 2025: Codility now analyses how code was typed, not only whether it was pasted. Retyping an answer from a second screen is no longer a gap in the system — it is a signal the system is specifically built to catch.

What proctoring records when a recruiter enables it

Codility groups its behavioural proctoring into a small set of events that appear on the candidate report's Timeline and integrity widget. Its assessment integrity documentation lists the following:

  • Paste events into the IDE, with the pasted code itself logged in the Timeline.
  • Tab switches away from the Codility browser tab.
  • Unusually fast completion relative to the expected time for the task.
  • Attempts to copy the task description, which Codility describes as "a common signal of AI tool use."
  • Webcam snapshots taken at intervals and when a proctoring event fires, retained for 30 days.
  • Optional continuous screen recording, and video recording of screen, webcam, and microphone for the whole session.

Two details matter more than the list. First, proctoring is configured per test and, per Codility's documentation, "must be enabled before the first candidate invite" — so whatever is running was decided before you opened the link. Second, when webcam or recording is enabled you are asked to consent: Codility states candidates "must select 'Allow' in order to proceed." If you are not shown a camera permission prompt, that particular layer is not running.

Codility is explicit that these events are not verdicts. Its behavioural events documentation states that "not all behavioral signals, on their own, are indicative of cheating," and its integrity page tells recruiters to use them to decide "whether a Codility Interview follow-up is warranted." The paste signal behaves much like the paste tracking used across other assessment platforms: it records the event and the pasted content, and leaves the interpretation to a person.

Typing Pattern Detection is aimed at retyping

This is the feature that closes the obvious workaround. Codility's Typing Pattern Detection documentation describes it as "a passive behavioural proctoring signal designed to identify candidates who may be retyping solutions generated by external AI coding assistants."

The mechanics Codility documents are worth reading precisely:

  • Analysis happens after submission, not live, and looks at "the sequence and structure of keystrokes" to judge whether a solution "was written organically or in an unnatural coding pattern consistent with copying from an external source."
  • A risk flag then appears in the Screen report next to the other behavioural signals.
  • It is on by default and can be turned off per assessment.
  • Solutions under 10 lines are skipped automatically "to reduce the potential for false positives."
  • It does not apply to multiple-choice, debugging, essay, file upload, or VS Code tasks.
  • Codility states the flag indicates "a high likelihood, not certainty, of retyping" and should "not be used alone as grounds for automatic candidate rejection."

The practical reading for a candidate is direct. Organic coding leaves a messy trace: false starts, edits above a line already written, a variable renamed halfway through, blocks deleted and rebuilt. Transcription is linear and clean. That difference is what the model is looking for, and typing a solution slowly does not make it iterative. Note also the internal inconsistency in Codility's own material as of 3 September 2026 — the behavioural events page still lists typing pattern as "Currently in Preview" while the dedicated page describes it as on by default — which is a reasonable reminder that documentation lags rollout in both directions.

Device Integrity requires installing the Codility App

Codility's newest layer moves the assessment out of the browser. Under Device Integrity for Screen and the interview equivalent, both marked Preview and last updated in mid-2026, "the candidate must download and install the Codility App in order to take the assessment," and the feature "identifies hidden cheating tools running on a candidate's device."

What Codility reports to the recruiter is narrow: whether the app was used at all, and "whether any external tools were detected during the session." What Codility says it does not do is equally specific — the app "does not record the candidate's screen, monitor keystrokes, or collect personal data beyond what the standard Codility assessment already collects." The app runs on macOS and Windows; "Linux is not currently supported."

Treat "detects known cheating tools" as a vendor claim about a Preview feature, not a measured capability. Codility does not publish the detection list, the method, or a false-negative rate, and no independent evaluation of it exists that we could find. The defensible conclusion is narrower and still decisive: if a Codility assessment asks you to install a desktop application before you can start, the employer has chosen a format where running other software during the session is squarely against the rules, whatever any given tool does or does not evade. That is a policy fact, and it does not depend on the detector working.

The older layer: Similarity Check and IP Check

Two long-standing signals sit underneath all of this and are frequently mistaken for AI detection.

Similarity Check compares your submission against Codility's corpus of prior submissions — the documentation cites "over 12 million assessments" — plus leaked solutions scraped from the web, and it is designed to recognise matches "even if solutions have been modified (e.g., changing identifiers, reformatting code, or making small changes in the code structure)." Codility tells recruiters plainly that "the warning doesn't necessarily suggest cheating," noting that simple tasks produce coincidental matches and that one candidate taking the same test for several companies can match themselves.

IP Check, documented by Codility since 2021, flags multiple devices or locations during a test and surfaces it under Suspicious Activity. Codility's caveat is the same: a flag "is not a guarantee that the candidate has cheated," since a home connection dropping to mobile data produces one.

Neither of these targets AI. A model-generated solution to an uncommon task is often less similar to prior submissions than a memorised idiomatic one.

What a flag actually triggers

Nothing on this list auto-rejects you. Every Codility document quoted above pushes the decision to a human, and the platform now supplies a specific follow-up mechanism: AI Follow-Up Questions generate three questions from your submitted code, each "time-boxed to 2 minutes, with copy-paste prevention and keystroke monitoring active during answers." Elsewhere the recommendation is an interview where you explain "how they came up with this solution, their train of thought."

That is the real consequence, and it is the same across most platforms: a flag converts a written test into a verbal defence of your own code. The general shape of that process — who sees the flag, what you are typically told, and what a follow-up looks like — is covered in what happens when a coding assessment flags you. Codility's stack is stricter than most on typing behaviour and looser than HackerRank's on live environment lockdown; for that comparison, see what HackerRank proctoring detects and records.

What to do before you start a Codility test

  1. Read the invitation and the consent screen. They tell you whether webcam, recording, or the Codility App are in play. An assessment that never asks for camera permission is not silently recording video.
  2. Ask the recruiter what is permitted, in writing, if AI use is not addressed. Availability of a tool is never permission; the question of when AI is actually allowed in a technical interview is answered by the employer's policy, not by what the platform fails to block.
  3. Do not treat retyping as a workaround. As of 2026 it is the specific behaviour Typing Pattern Detection was built to flag, and it applies by default.
  4. Expect to explain your code. Whether or not a flag fires, the follow-up question is the norm on Codility now. A solution you cannot walk through line by line is a liability independent of any detector.

Where a desktop assistant fits, and where it does not

Control is a desktop AI interview assistant for Windows and macOS that runs as a native overlay rather than a browser extension, transcribes the interviewer in real time, accepts screenshots of a problem, and can be driven from a paired phone remote. Its screen-capture behaviour is scoped rather than absolute: Control states it stays out of supported screen captures and does not take focus from the shared window, with the supported set named on its site. New users get 5 messages and 2 minutes of voice free; passes are $12 for 24 hours, $19 for 72 hours, or $39 monthly, as listed on 3 September 2026.

Those characteristics are relevant to a live screen-share interview where the rules permit assistance. They are not an answer to a proctored Codility Screen with Device Integrity enabled, and we would not present them as one — that is a closed-book assessment under an installed client, and the honest advice is to take it closed-book. If your interviews are the live, screen-shared kind instead, the screen-share workflow guide is the more useful next read.

The short version

A Codility test in 2026 is a closed-book assessment with post-hoc behavioural analysis, and the analysis now covers how the code arrived on the page. Paste logging, tab switching, completion speed, task-description copying, keystroke sequence, similarity, and — increasingly — an installed client are all available to the employer, but which ones are running was decided before you were invited. Read the consent screen to find out what you are facing, ask about AI policy rather than inferring it, and prepare to explain your solution out loud. That last habit survives every change to the detection stack.

Specific Codility questions

Continue exploring

Control AI - What Codility Detects in 2026: Proctoring and Typing Patterns