Fabric’s Cheating Data Argues Against the Interview It Sells

Abhishek Vijayvergiya, co-founder of Fabric, tells a story that has become routine for him. A CHRO at a large IT services company in Mumbai watched a candidate clear every virtual interview round and land on a client-facing project. Within a week, the client escalated. “The candidate completely failed on the customer project and showed no skills that he was rated highly for during the interview,” Vijayvergiya said. “I’m hearing similar stories in almost every customer conversation.”

Fabric, an agentic AI hiring platform whose own agents conduct those Round 1 interviews, has published an analysis of 19,368 Round 1 sessions, first reported by Ahmedabad Mirror, to explain why. Its prescription is more detection built into its own platform. Read its own numbers plainly, and they argue for something else: no amount of detection fixes an interview that never had a person in the room asking a real question in real time.

What Fabric’s Own Numbers Show

These are Fabric’s figures, from Fabric’s platform, describing Fabric’s product. Treat them as a vendor’s account of its own blind spot rather than an independent audit.

Across 19,368 Round 1 interviews, Fabric flagged 7,457 candidates, 38.5% of the total, for AI-assisted cheating behavior. More than 61% of those flagged candidates still cleared the platform’s own pass bar of 7 or higher out of 10. Without the detection layer, Fabric says, those candidates would simply have advanced to the next round.

The pattern was not even. Technical roles were flagged at 48%, against 12% for sales roles, and Fabric’s report does not explain why technical roles skew higher. It only reports that they do. Junior candidates, those with zero to five years of experience, were flagged at nearly double the rate of more experienced hires.

The 82% No Proctoring Tool Catches

Fabric’s breakdown of method matters more than the headline rate. Of the flagged behavior:

• 45% used a dedicated interview assistant such as Cluely, a $50 to $150 monthly subscription.
• 34% used ChatGPT voice mode or a similar general-purpose model.
• 18% used tab switching, the old trick.
• 3% had live help from someone off camera.

Tab switching is the only method most proctoring tools were built to catch. That leaves 82% running quietly in the background, invisible to the detection most hiring teams already have in place. This is not a story about candidates getting cleverer at hiding. It is a story about the format of the interview rewarding whoever produces the smoothest answer, regardless of where that answer came from.

Why More Detection Won’t Close the Gap

Fabric’s answer to its own findings is 21+ additional signals, timestamped evidence, and a human who reviews the flag before deciding. That is a reasonable safeguard for the decision. It does nothing for the interview itself, because the interview being gamed is a fully automated Round 1: an AI agent asks the questions, a candidate answers, and no person listens live to push back.

In our work with hiring teams, the failure point is rarely the first answer. It is the second question, the one that asks a candidate to build on what they just said. A scripted answer only has to survive a script. It does not survive a real question asked by someone who is actually listening. An interview format with no live human asking that second question measures fluency, not skill. That is exactly what an LLM running in the background is built to supply.

The Fix Was Already in the Report

Fabric’s own recommendations list three tactics: watch behavioral patterns, dig deeper with personal follow-up questions, and use AI detection tooling. Only one of those three fixes the interview instead of auditing it after the fact. Watching for delays or repeated phrasing tells a recruiter that a candidate might have had help. A real follow-up question, asked live, tells the recruiter whether the candidate can actually think.

Fabric’s own report notes that AI-generated answers become vague, repetitive, or disconnected under follow-up. That description matches what a skilled human interviewer already does. It is an interviewing skill, not a technology purchase.

This matters most where Fabric’s data says the risk is highest. Junior candidates were flagged at nearly double the rate of experienced hires, and junior and campus funnels run the fewest interview rounds. A coached answer that clears an automated Round 1 in that funnel does not get a second chance to be caught. It becomes the hire.

What VPs of TA Should Actually Fund

If your highest-volume, lowest-experience funnel is also the one with the thinnest human contact, ask what that funnel is actually built to measure. The instinct when a vendor reports a 38.5% flag rate is to buy a better flag. The harder instinct is to put a trained interviewer back into the round where volume pressure took them out, and fund the skill of asking a follow-up question that a rehearsed answer cannot survive twice. It costs more up front than another layer of detection. It holds up for longer.

Original reporting: Ahmedabad Mirror.