DeepMind and GovTech Build Scam-Spotting AI for the Job Offer That Wants to Be Paid First
The Singapore collaboration uses an on-device Gemma model to flag manipulative conversations. That wonderfully generous stranger offering work may face a less enthusiastic reader.
Automation ·
A job offer arrives without an interview. The pay is excellent. Then comes a small payment from you to get things moving. An unusual interpretation of payroll, and apparently one worth teaching a phone to question.
Google DeepMind and Singapore's GovTech announced their scam-detection collaboration on October 14. They have fine-tuned a Gemma model on local scam conversations, with fake job offers among the scams discussed in the announcement. The planned application rollout will be phased. With user consent, the model is designed to analyze conversations on the device and warn when it detects manipulation, without sending those conversations to Google or the government through the deployment.
The proposed system starts with a lightweight screening model. Suspicious material gets a closer look using more of the conversation. I approve of reading beyond the flattering introduction. A stranger can be extremely impressed by your potential while remaining strangely interested in your bank balance.
There is no published detection-rate figure in the announcement, so the charming fake recruiter has not been declared extinct. But the sales pitch may eventually have to survive a second reader on the same phone. This one is less concerned about missing a once-in-a-lifetime opportunity and more interested in why the employer has sent the applicant an invoice.
Based on: Google DeepMind and Singapore Government collaborate to build fine-tuned AI model for scam detection, Google DeepMind, October 14, 2026.