June 15, 2025
Why Goldman Sachs’ ChatGPT Clamp-down Signals a New Era of Interview Integrity
Goldman Sachs has joined Amazon and Anthropic in telling candidates “leave Gen AI at the (virtual) door.” Their message is clear: if your applicants outsource answers to ChatGPT, you can’t gauge true ability or culture fit.
News
6 Min Read
TL;DR - Goldman Sachs has joined Amazon and Anthropic in telling candidates “leave Gen AI at the (virtual) door.” Their message is clear: if your applicants outsource answers to ChatGPT, you can’t gauge true ability or culture fit. Below is a step-by-step framework—and a free compliance checklist—to help talent teams detect AI-assisted answers, cut interview fraud, and keep time-to-hire on track.
The Crackdown Is Real
Goldman Sachs now warns student applicants: “please do not use AI assistants during the interview process.” (Fortune)
Amazon tells recruits they may be disqualified if caught using “Gen AI tools” live. (Business Insider)
Anthropic echoes the stance, stressing that it wants “non-AI-assisted communication skills.”
Bottom line: Fortune-500 employers see candidate cheating in coding interview and behavioral screens as a rising risk to productivity and trust.
Why HR Can’t Ignore the “AI Teleprompter” Problem
False positives create churn. If you miss AI-assisted answers and a hire under-performs, you’re back to square one 3-6 months later.
Brand reputation. New-grad forums already trade tips on “real-time keystroke analysis interview” dodge tactics.
Legal exposure. Using external AI can violate NDA-level confidentiality clauses embedded in many interview tasks.
4 High-Signal Behaviors That Flag ChatGPT-Assisted Responses
Signal | What it looks like in Zoom/Teams | Tooling tip |
|---|---|---|
Eye-tracking drift | Eyes slide sharply side-to-side as if reading a second screen | Enable gaze-vector alerts in your AI-powered integrity platform |
Head-down typing | Audible keyboard bursts mid-answer | Use real-time keystroke analytics plus on-screen keyboard logging (no content stored) |
Latency spikes | 3–7 second pauses before perfect reply | Set a 2 second thinking-time threshold in live proctoring |
Off-topic eloquence | Fluent response that never quite answers the asked question | Run semantic overlap scoring between prompt & answer |
4-Step Remediation Framework
H2O2 - a quick mnemonic your team can adopt this quarter.
Harden settings
Lock browser tabs; deploy screen-share monitoring for coding tests.
Harvest signals
Combine video gaze, keystrokes, and clipboard events in a single dashboard.
Observe anomalies
Benchmark each metric against a human-only baselined cohort.
Optimize & educate
Share anonymized clips in interviewer training; update policy wording every 6 months to stay deterrent-focused.
Try This in Your Next Round
Add a single line to your invite: “Live AI assistants prohibited unless stated; sessions are monitored for policy compliance.”
Open with an ice-breaker question you didn’t include in pre-read material—proxy interviewer fraud signs show up fastest here.
Use Proctaroo’s behavioral real-time detection for online assessments and interviews.
One-Click Next Step
Ready to see how live proctoring vs record-and-review stacks up in your stack? Book a 15-minute demo and cut interview fraud before your next campus recruiting wave hits.
