Short answer
Not reliably. Australia's higher education regulator, TEQSA, notes that AI use in unsupervised assessments appears almost impossible to detect reliably. Many interview assistants are built to stay out of a screen share, and help can come from a second device the software never sees. Controlling the room is more dependable than trying to spot AI afterwards.
Key takeaways
- Detection software can only watch the computer it runs on, while most AI help now comes from overlays, voice assistants and second devices.
- In Fabric's data, 79% of flagged AI use came from tools designed to be undetectable by screen sharing.
- Automated flags can misread honest behaviour such as looking away to think, which is unfair to good candidates.
- A parliamentary committee has recommended AI used in recruitment be classed as high-risk, raising the bar for defensible hiring decisions.
Why is AI use in interviews so hard to detect?
Because the help usually doesn't run where the detection software is looking. A remote interview shows the interviewer one camera angle and, at most, one shared screen. AI assistance can sit outside both.
- Overlay apps draw answers on top of the screen but are excluded from screen sharing.
- Voice mode on a chatbot listens to the question and replies through an earbud.
- A phone or second laptop just below the camera shows answers in real time.
- Another person can relay answers from off-camera.
Fabric's analysis of 19,368 interviews found dedicated interview assistants (45%) and chatbot voice mode (34%) made up 79% of flagged cases. Both are designed to stay invisible to screen sharing.
What do Australian regulators say about AI detection?
Australia's higher education regulator has been direct about the limits. In guidance published by TEQSA, the authors write that it "appears almost impossible to detect" generative AI use in assessment work in a reliable way, and that it's hard to design unsupervised tasks that rule out substantial AI use.
That guidance was written for universities, but the logic carries over to hiring. A take-home task or an unsupervised video interview has the same weakness, because nobody can see the whole room.
Parliament is also paying attention to AI in hiring. The House of Representatives Standing Committee on Employment, Education and Training recommended that AI systems used for employment purposes, including recruitment and hiring, be classified as high-risk.
What do proctoring and detection tools actually check?
Most tools monitor the candidate's own computer and webcam. Typical checks include browser lockdown, tab-switch tracking, webcam monitoring, eye or head movement analysis and pattern analysis on answers.
Each check has the same blind spot. None of them can confirm what's happening on a phone, in an earbud, or just outside the camera's view. Proctoring software also doesn't verify identity against government ID in person.
Can detection tools wrongly flag honest candidates?
Yes, and that's a fairness problem as much as an accuracy one. Candidates look away to think, read the question twice, pause on a slow connection or get nervous. Automated systems can read any of these as suspicious.
When a flag appears, the employer usually has no way to prove what actually happened. An honest candidate can lose an opportunity, and the employer can lose a good hire, on the strength of a guess.
How do the main interview options compare?
| Remote, unsupervised | Remote with proctoring software | In person at your office | Supervised pod | |
|---|---|---|---|---|
| Blocks AI tools and second devices | No | Partly | Yes | Yes |
| Checks government ID | No | No | Yes | Yes |
| Same conditions for every candidate | No | No | Yes | Yes |
| Protects honest candidates from false flags | No | No | Yes | Yes |
| Candidate travel | None | None | To your office, often interstate | To a local venue |
| Written record of the session | No | Partly | No | Integrity report |
What works better than detection?
Removing the opportunity works better than trying to catch it. In a supervised setting, devices go into a locker, the computer runs only the interview platform and a person is present for the whole session. There's nothing to detect because there's nowhere for an AI tool to run.
In-person interviews achieve this, but they're expensive when candidates are interstate. A supervised interview pod gives the same controlled conditions at a local venue while your panel joins remotely. Read what a supervised interview pod is and how it works, or see the latest figures on AI use in interviews.
Frequently asked questions
Do AI interview detection tools work?
Only partly. They monitor the candidate's own computer and webcam, so they can miss overlays built to hide from screen sharing, chatbot voice mode through an earbud, and second devices out of view.
Is proctoring software the same as supervision?
No. Proctoring software watches a candidate's computer remotely and flags behaviour for review. Supervision means a person is present in the room, devices are secured and the computer is locked to the interview platform.
Does SmartVue use AI detection software?
No. SmartVue focuses on the room instead. Staff check government ID, phones and smartwatches go into lockers, the pod computer runs only your interview platform and every session is supervised.
Can I keep using my own interview platform?
Yes. Supervised pods work with Zoom, Microsoft Teams, Google Meet, HackerRank, CoderPad or your own assessment tools, and your panel interviews as usual.
Sources
- TEQSA, "Assessment reform for the age of artificial intelligence", 2023
- Parliament of Australia, House Standing Committee on Employment, Education and Training, The Future of Work, list of recommendations
- Fabric, "State of AI Interview Cheating in 2026: Insights from 19,368 Interviews", 14 January 2026
