Singapore scam-detection AI announced: an on-device model, with app rollout still phased

Google DeepMind’s October 14 announcement describes a model with GovTech Singapore. It does not establish that a protection app is already enabled on your phone.

The partners adapted Gemma for local scam conversations. DeepMind says app deployment will be phased, with access to on-device conversations requiring the user’s consent under the application’s terms. These are the developer’s statements, not our deployment or privacy audit.

A model announcement is not your installation

Our interpretation: separate the underlying model from the application that would use it. Before relying on a feature, look for the actual app’s official availability, supported device and consent information. We did not verify access for an individual phone or for users in Korea.

Keep the claim about processing specific

The stated design processes conversations on the device. That alone does not verify every other function of a future host app. Review its published handling of messages and permissions rather than extending one model claim to the entire phone.

A missing alert cannot certify a conversation

We did not measure missed scams or false alarms. Our recommendation is to keep your usual independent verification before sending money or credentials. An announcement, or an absent warning, is not evidence that a particular request is legitimate.

Published October 14, 2026. The official announcement is dated October 14; an individual app’s public deployment date was not established here. No app, private conversation, consent flow, classifier or scam interaction was tested.

Sources: Google DeepMind · Scam detection.