Google DeepMind and GovTech unveil on-device scam detection model – Unite.AI

Google DeepMind and the Government Technology Agency of Singapore (GovTech) announced on October 14, 2026, an on-device scam detection model, created to alert smartphone users to manipulative conversations without sending private messages to the cloud, in a country where scams have caused more than $4 billion in losses since 2019.

Losses due to scams and the move to social engineering

According to the announcement on the Google DeepMind blog, written by JJ Geewax and Miguel de Andrés, scams account for more than half of reported crimes in Singapore and have caused losses of more than $4 billion since 2019. The announcement states that scam tactics have grown from simple phishing to complex social engineering techniques, in which attackers construct convincing conversations to manipulate victims before asking for money, and that such scams now account for up to 80% of all losses.

Generic models often struggle with localized deception techniques, making scams more difficult to detect, the announcement said. Google DeepMind’s lightweight Gemma model was tuned to local scam conversations, ranging from impersonations of public institutions, banks and local figures to trending regional schemes. The announcement states that this allows the model to distinguish harmless social chatter from real manipulation, and that the resulting model is optimized to run on typical Singaporean smartphones.

On-device processing and two-step pipeline

The model processes conversation data on the device itself. Conventional threat detection scans messages on servers in the cloud, the announcement explains, but many messaging platforms are end-to-end encrypted and therefore cannot inspect private conversations. Performing analytics locally allows for faster examination of conversation patterns to detect manipulative intent, so people can take action before financial or credential exposure occurs.

When the model detects manipulative conversation patterns, a real-time alert appears on the device, urging the user to take a break before sharing sensitive information or money.

According to the announcement, battery efficiency was a key concern in running a complex scam classifier on modern smartphones. The teams approached this problem with a two-stage pipeline. In Phase 1, a lightweight and efficient classifier filters incoming messages using minimal power, focusing on identifying potential threats. In Phase 2, the optimized Gemma model performs a contextual evaluation of a larger snapshot of the conversation to confirm a threat or discard a false alarm.

Governance measures

The collaboration was showcased at the Singapore International Cyber ​​Week 2026. The event’s keynote speech was delivered on 13 October 2026 by K Shanmugam, Senior Minister and Minister of Home Affairs of Singapore, according to the speech published by the Ministry of Home Affairs.

The announcement describes rigorous governance maintained throughout development under measures labeled Sovereign Sandbox, No Citizen Data, and Stress-Testing. As part of the Sovereign Sandbox, tuning occurred within the Google Cloud Platform environment managed and maintained by GovTech, keeping training data isolated. With No Citizen Data, the training strictly excluded private or personally identifiable information and used anonymized datasets drawn from scam reports and intelligence feeds. Under stress testing, the model was tested against synthetic dialogues and local corpora covering different scam categories and evolving imitation tactics.

“Fighting scams shouldn’t mean sacrificing personal privacy,” said Mark Chen, Director Trust and Safety at GovTech Singapore. “Working with Google DeepMind, we are bringing advanced AI protection directly to citizens’ devices. This allows us to spot highly targeted local scam attempts in real time, while keeping personal communications completely private.”

Background to the National Partnership on Artificial Intelligence

The scam detection model was provided as part of Google’s national AI partnership with the Singapore government. Google DeepMind announced this partnership on May 20, 2026, launching new programs in the country under the National Partnerships for AI initiative after opening a research lab in Singapore the previous year.

Such programs include an exploration of collaboration of AI co-clinicians with public health clusters; supporting infectious disease research and pandemic preparedness with Google.org, part of a $7 million contribution to the Philanthropy Asia Alliance’s Health for Human Potential coalition; a Gemma-based running assistant for blind and visually impaired athletes, tested and retested with SG Enable, Singapore’s focal agency for disability and inclusion; training for local researchers on artificial intelligence scientific tools with the National Research Foundation; Gemini for Education was provided to educators from primary schools to middle schools; and multimodal and multilingual security benchmark research with Infocomm Media Development Authority and MLCommons.

Integration path for messaging providers and operating systems

The announcement states that messaging platforms and operating system vendors can directly integrate these lightweight models to safeguard users locally, keeping chat interactions secure without exposing private data. He presents the project as a model for messaging platforms that balance user protection with end-to-end encryption, and as an example of governments and model vendors working together to build technology for the public good.