Ghost Coders and AI Proctors: How Hiring Fraud Is Reshaping Campus Recruitment

Picture a candidate seated alone in a quiet room, webcam on, timer counting down, fingers moving steadily across a keyboard. On the surface, the scene looks like any standard technical assessment. But behind the screen, a remote operator may be silently solving the problem through desktop-sharing software, while a secondary device feeds AI-generated solutions off-camera, and a wearable earpiece relays instructions from somewhere else entirely. The candidate is present. The candidate, in any meaningful sense, is not.

This is the new frontier of hiring fraud in India’s technology sector, and it is growing more organised by the month. What began as opportunistic cheating — a copied snippet here, a whispered answer there — has matured into what recruiters now describe as structured “interview-as-a-service” networks, coordinated primarily through Discord and Telegram groups, where proxy candidates and ghost coders are available for hire at scale. The rise of generative AI has not created this problem, but it has industrialised it, lowering the technical barrier to fraud while raising the stakes for every company that relies on campus hiring pipelines to source engineering talent.

The Anatomy of a Compromised Assessment

Vikas Aditya, CEO of HackerEarth, identifies four dominant forms of malpractice currently circulating through the hiring ecosystem. The first is AI-generated code submissions, where candidates feed the problem statement directly into a large language model and submit the output with minimal modification. The second is the deployment of proxy candidates — individuals hired specifically to sit assessments on behalf of the actual applicant. The third involves off-camera assistance, either from a second device running an AI tool or from another person physically present but out of frame. The fourth, and perhaps the most technically sophisticated, involves virtual machines or remote-desktop software that conceal an entirely separate AI session from the proctoring system monitoring the primary screen.

What unites these methods, Aditya argues, is that they all exploit the same structural weakness in conventional assessment design. “The common thread is that nearly all of these exploit the same weakness — a static assessment that scores only the final answer without observing how it was produced,” he said. A system that measures only outcomes, without interrogating the process that generated them, is, by design, blind to the question of authorship. The final score looks identical whether a candidate spent forty minutes working through a problem or spent forty seconds pasting it into ChatGPT.

Proctoring at Scale

The industry’s primary institutional response has been the rapid adoption of AI-powered proctoring platforms. Companies including many of India’s largest technology employers now deploy tools such as Talview and Mercer Mettl alongside assessment platforms like HackerRank and HackerEarth, using systems that analyse webcam feeds, audio streams, and screen activity simultaneously to flag anomalous behaviour in real time. The shift is also a structural one: firms are moving away from browser-based assessments, which are more permeable, toward locked-down environments that restrict what a candidate can access during an active session.

The data reflects how quickly this transition is occurring. HackerEarth’s 2025 Technical Hiring Landscape Report found that the share of companies using proctored technical assessments rose from 64 percent at the start of the year to a peak of 77 percent by July, with nearly two-thirds of all technical assessments conducted over the year falling under some form of proctoring. That trajectory is steep, and it suggests that companies which had previously treated assessment integrity as a secondary concern are now treating it as a primary operational risk.

Vivek Ravisankar, co-founder and CEO of HackerRank, offers a figure that puts the scale of the problem in sharper relief. “We flag about 30 to 35 percent of sessions with at least one suspicious behaviour,” he told the Times of India. “The biggest issue is the use of AI-powered cheating apps.” That is not a marginal anomaly rate — it is, by any measure, a systemic signal that the integrity of unproctored or lightly monitored assessments is now genuinely compromised across a significant proportion of candidates.

The Human Check That Machines Cannot Replace

For all the sophistication of automated proctoring, HackerEarth’s most effective countermeasure turns out to be disarmingly low-tech. Aditya describes a short live follow-up interview in which candidates are asked to explain their submitted solution in their own words, walking through the logic, the trade-offs, and the reasoning behind specific implementation choices. The results are telling. “Most candidates who relied on AI fail within two questions,” he said — a finding that speaks both to the effectiveness of the intervention and to the superficiality of the preparation it exposes.

This points to a deeper structural issue that the proctoring boom alone cannot resolve. The industry, as HackerEarth’s report acknowledges, likely underestimates the true scale of AI-enabled cheating precisely because conventional coding tests surface only a final score, revealing nothing about the cognitive process — or absence of one — that produced it. A candidate who cannot explain a solution they have submitted has not demonstrated competence; they have demonstrated the ability to operate a prompt interface. These are not equivalent skills, and the gap between them is exactly what ghost coders and AI tools are currently exploiting.

The recruiters deploying proctoring platforms and live verification interviews are, in effect, attempting to reintroduce process visibility into an evaluation architecture that had stripped it away in the name of efficiency and scale. The irony is that the very tools which made mass campus hiring operationally feasible — standardised online assessments, automated scoring, asynchronous scheduling — are the same tools that made it structurally vulnerable. Fixing the vulnerability, it turns out, requires reintroducing the human element that efficiency once displaced.