Google DeepMind and Singapore's Government Technology Agency (GovTech) announced on October 14 that they have fine-tuned a lightweight Gemma model to detect locally relevant scam conversations. The project targets investment fraud, fake job offers and impersonation schemes that rely on sustained social engineering.
Phased app integrations, with consent
Google describes a two-stage process: a lightweight classifier screens incoming messages, and the fine-tuned Gemma model reviews suspicious conversations for manipulative patterns. The companies say any application rollout will be phased and that on-device processing would require the user's consent under the application's terms.
This is a development and partnership announcement, not confirmation of a broad public launch. The privacy and detection claims are described by Google DeepMind; no independent accuracy evaluation or deployed-app test was included in the announcement.