AI posture: on or off, fully reviewable_
Every assessment carries an AI posture your account sets in advance. Turn the AI Assistant on, or turn it off. Whatever happens next is captured and visible to a human reviewer, not scored automatically. The decision about what it means stays with you.
The candidate works with the AI Assistant available. Every interaction is captured as reviewable AI activity for the reviewer to see, not a hidden score.
The AI Assistant is unavailable. Integrity’s layered controls, identity, behavior, similarity, protect the signal a human reviews.
Every AI policy claim needs to hold up when someone asks how it works
Talent teams are under pressure to have an AI policy for hiring. Engineering leaders want to know candidates can actually build. Regulators want to know a machine is not making the hiring decision. A vague answer does not survive any of those conversations.
A blanket AI ban hides real capability
Blocking AI everywhere loses signal on candidates who need to collaborate with AI on the job. You end up assessing a workflow that no longer matches how the role is actually done.
Unlogged AI access hides real skill
Allowing AI tools with no visibility into how they were used tells you a candidate produced working code. It does not tell you whether they understood it or delegated the whole task.
Regulators are asking who decides
The EU AI Act and adjacent frameworks want a clear answer on whether AI or a machine made the hiring call. A platform that cannot answer that plainly is a liability.
Candidates challenge results without evidence
A pass or fail score with nothing behind it invites disputes. A reviewer needs to see what actually happened in the session, not just a number.
One switch per assessment. Every interaction stays visible
Your account sets whether the AI Assistant is available for a given assessment. There is no separate setting per skill, per question, or per candidate inside that assessment. Whichever posture you choose, the record is the same: fully captured, fully reviewable, reviewed by a person.
Pure skill measurement
Use this posture where you want to see what a candidate can build without an AI Assistant in the environment.
- The AI Assistant is unavailable to the candidate for the full session
- Integrity’s layered controls run in the background: identity, behavior, similarity, risk scoring
- A reviewer sees the full integrity record alongside the submission, not a pass or fail alone
This posture does not detect every possible workaround. It gives a reviewer the evidence to make the call.
Working with AI, on the record
Use this posture where working alongside AI is part of the job you are hiring for.
- The AI Assistant is available to the candidate for the full session
- Every prompt, suggestion, and follow-up is captured as reviewable AI activity
- None of that activity affects automated scoring. A reviewer reads it and decides what it shows
Transparency is the point here, not restriction. You see how the candidate worked, not just what they produced.
Where this is headed: model-level and more granular AI controls are on the roadmap. Today, the posture is set once per assessment, not per skill, per question, or per role.
When AI is off, Integrity is what protects the signal
Turning the AI Assistant off does not mean turning off scrutiny. Integrity layers five kinds of evidence into one record for every AI-off session. No single signal decides anything on its own, and none of it auto-rejects a candidate.
-
ID verification
Candidates confirm identity with a government-issued ID before the assessment begins, so the person you review is the person who sat the session.
-
Behavioral monitoring
Tab switching, time away from the environment, and paste volume are tracked through the session and surfaced to the reviewer, not judged automatically.
-
Cheating apps detection
Surfaces invisible AI helper tools running alongside the assessment window, tools designed not to show up on a shared screen.
-
Similarity check
Submissions are compared against known and leaked solutions. A match is flagged for a reviewer, it does not reject a candidate on its own.
-
Qualitative risk scoring
Every signal rolls up into one deterministic risk level: none, low, moderate, or high. The scoring is rule-based, not a generative model guessing at intent.
-
Human review, always
A risk level is a prompt to look closer, not a verdict. A person reads the evidence and makes the call on every flagged session.
Risk levels help a reviewer prioritize what to look at first. They do not auto-reject a candidate, and they are not a claim of certainty about what happened.
From a vague policy to a plain answer
Without a clear posture
With Codility
No AI in the hiring decision, stated plainly
Codility’s posture under the EU AI Act is explicit: no AI or machine learning makes an automated hiring decision. AI activity is captured for a human to review. Integrity risk levels are deterministic, not a model’s guess. The decision about a candidate is always a person’s decision.
How it works
Set the posture for the assessment
Your account decides, ahead of time, whether the AI Assistant is on or off for that assessment. The setting applies to the whole session.
The candidate completes the session
Integrity signals or AI activity, whichever applies, are captured automatically. Nothing about the record disrupts the candidate experience.
A reviewer reads the evidence
The integrity risk level or the AI activity log sits alongside the submission. A person decides what it means. Nothing here rejects a candidate on its own.
Technical hiring and skills assessment