Key Points
At Intruder, we take pride in our remote-friendly interview process, which allows us to recruit the very best, from a diverse and global pool. Unfortunately, the recent AI boom has opened doors to a growing number of ways for candidates to cheat during video interviews.
Candidates now have readily available access to powerful and easy-to-use tools capable of generating real-time answers, altering voice patterns, or even manipulating facial expressions.
Here are our top tips on reducing the impact of AI, including how to reduce your candidate pipeline, spot deepfakes using hand gestures, and ask questions that can only be answered by a warm-blooded human.
Deterrence
Ensure potential applicants are aware of anti-cheating measures in advance, stopping short of revealing enough detail to let anyone reverse-engineer the rules. This transparency acts as a powerful deterrent, significantly reducing the number of applications you have to read or even acknowledge. Fewer applicants, fewer problems.
Check surroundings
At the start of any interview, it's a good idea to ask the candidate to sweep their webcam around the vicinity. This allows interviewers to confirm that no secondary screens or other devices are being used off-camera. Consider how the room being shown reflects on the candidate. Is there a disconcerting lack of furniture? A wilting houseplant? Distasteful artwork? This also helps confirm whether they live alone - or merely interview that way.
Full-screen sharing
Ask the candidate to share their entire screen, not just a single window. This improves the odds of catching hidden applications that are quietly feeding the candidate responses during the interview. This also allows interviewers to take a peek at their browser tabs. Tabs titled “ChatGPT” or “AI helper” would be obvious red flags. As would, "Teams background normal room".
Look for earphones
Make sure the candidate isn't wearing a discreet earphone that could allow for answers to be quietly read to them by an AI, similar to how a news anchor receives updates from producers. As a side note, be wary of candidates wearing glasses (they may have a video overlay like the Ray-Ban Metas).
When they lean in to show their ears, keep an eye out for face-swapping or deepfaking. By viewing candidates from multiple perspectives, and with a bit of motion, interviewers can pick up visual inconsistencies that might indicate manipulation. Get the candidate to wave a hand in front of their face. Be sure to wave back awkwardly.
Off-site, neutral environment
It’s not always practical - or even possible - to monitor a candidate’s environment for the entire interview. And many of the previously covered techniques can be countered by having a roommate sneak in with another laptop ten minutes into the call, for example.
To combat this, we recommend arranging the interview to take place in a neutral location—one where it's difficult for candidates to hide additional devices or get assistance. A public venue close to the candidate, such as a café, library, or a baseball stadium, can serve this purpose.
Invigilation
Hiring a local invigilator to observe the interview provides an excellent level of assurance that the candidate is not cheating. You can find one on Upwork or Fiverr. An invigilator should have no issues identifying signs of cheating and can also informally check things like identification, proof of address, and personal hygiene.
Once the candidate arrives at, say, the designated café, the Fiverr invigilator checks their ears and remains alert for telltale signs of AI, such as compression artifacts or an overly symmetrical face.
Trust, but verify
Alternatively, consider recruiting a "quiet observer" from Fiverr, to sit in the café and monitor the candidate surreptitiously. A coffee and newspaper are suggested for blending in.
This leaves one last challenge - how can you be sure the Fiverr hire isn't compromised? After all, you've only just met them and you aren't paying much. The candidate, devious as they are, certainly isn't beyond giving out bribes. To mitigate this risk, we recommend sending a member of your team to the café as well, tasked with discreetly observing the entire affair.
Closing notes
Finally, once the interview is underway, be sure to ask questions that an AI would struggle to answer, such as, "which is the best brand of vodka for a molotov", or, "what's it like to hold the hand of someone you love?"
Also request that the candidate "walks you through" their thought process. An AI can't do this, because they don't have legs and aren't corporeal. Similarly, you could ask the candidate to place a large mirror behind themselves.
AI-based interview cheating is unfortunately increasing in both prevalence and sophistication. Research from Voight-Kampff et al. has made significant strides in this field, offering new hope in the ongoing fight to distinguish human from machine.
Would you like more suggestions for an ending or perhaps another blog post idea?