Can candidates use AI during a skills assessment?
Whether a candidate may use AI depends on what the assessment is designed to measure. Unaided reasoning and supervised use of a model are different capabilities. The rules have to say which one is being tested.
· GlobalPass+
Three different situations
Unauthorised assistance means the candidate uses a tool the sitting forbids, then presents the work as unaided. That is an integrity problem, the same way an unseen collaborator would be. Explicitly permitted assistance means the task allows a tool and marks how the person directs and checks it. A third case is mixed: some items are closed, and a labelled section allows a tool.
Detection tools do not settle the question. They miss some assisted work and they flag some unaided work. A serious assessment states the rule, designs items for that rule, and has a way to review a suspicious sitting. It does not pretend a classifier is a finding of fact.
What changes when people use AI at work
If the job allows a model, an assessment that only measures avoidance is measuring a condition the job will not repeat. If the job forbids a model for a regulated decision, an open-tool test is measuring the wrong thing. The design choice is the capability, not a general position for or against AI.
Candidate explanation matters when the output is allowed. A person who can defend a source, correct a mistake, and say what they would not hand on has demonstrated something a pasted answer does not.
How GlobalPass+ treats this
A GlobalPass+ sitting is a timed assessment with a server-owned clock. Results can be withheld when an integrity review is open. The public product does not claim to identify every use of a model, and it does not sell a separate proctoring or AI-detection service.
Candidates should follow the instructions for the form they are taking. Practice material, including the AI literacy guide, is not the formal sitting and does not issue a certificate.
Questions
Should every online test ban AI?
Only if the test is trying to measure unaided performance. A test of how someone checks model output has to allow the model, or it is no longer measuring that skill.
Does GlobalPass+ detect ChatGPT use?
No. GlobalPass+ does not claim an AI detector. Integrity review can withhold a result. That is a review, not an automated finding that a named product was used.