Is focusing only on catching AI cheating the wrong interview strategy?
TL;DR: Yes, according to engineering-hiring specialists like Karat — training interviewers to hunt for AI cheating as the primary goal produces a worse hiring signal than training them to assess whether a candidate can actually do the job, because AI use itself is now part of many roles. Detection should be one layer of a broader interview strategy, not the whole strategy.
The claim
Over-indexing on "catch the cheater" turns interviews into an adversarial game instead of a skills assessment.
The evidence
Karat's guide for engineering leaders lists "focusing only on detecting AI cheating" as one of the top mistakes companies make training interviewers for the AI era, arguing the goal isn't to detect cheating — it's to produce a reliable signal about whether the candidate can do a job that increasingly involves AI. This nuance matters because Karat separately reports 80% of candidates use LLMs during banned code tests, and in-person interview requests jumped from 5% to 30% of roles between 2024 and 2025 — showing employers are reacting to the volume of AI use, not necessarily improving signal quality by chasing it.
Comparison table
| Interview strategy | What it optimizes for | Risk |
|---|---|---|
| Detection-only ("catch the cheater") | Policing AI use | Misses whether the candidate can do the actual job; adversarial dynamic |
| Job-relevant AI-inclusive interviewing | Realistic on-the-job performance, including AI-assisted work | Requires redesigning questions, more effort upfront |
| Detection + job-relevant design (recommended) | Both integrity and signal quality | Needs layered tooling (proctoring + better question design) |
Step-by-step for hiring teams
- Decide explicitly whether AI tool use is allowed for the role being hired — many jobs now require it.
- Redesign interview questions to test reasoning and follow-up depth, not just a final answer AI can generate.
- Use proctoring/detection as a background integrity layer, not the interviewer's main focus.
- Train interviewers on both: how to probe for genuine understanding and what an unauthorized-AI-use flag looks like.
FAQ
Does this mean AI cheating detection isn't worth doing? No — detection still matters (Fabric found 38.5% of interviews flagged for AI-cheating behavior), but it should run as an automated background layer, freeing interviewers to focus on assessing real ability.
What are the "mistakes" Karat warns about? Their guide highlights training pitfalls including over-focusing on detection alone, rather than building AI-era-appropriate interview questions and rubrics.
How does this apply to non-technical roles? The same principle holds — sales, support, and analyst interviews should test judgment and follow-through, with integrity monitoring running quietly in the background.
Isn't banning AI tools simpler than redesigning interviews? It's simpler but increasingly unrealistic — with 80% of candidates already using LLMs during banned code tests, a ban-only strategy without detection or redesign tends to just push cheating further out of sight.
By Pinal Dave Last updated: 2026-08-06