Google DeepMind and GovTech Singapore announced a collaboration to build an AI model that detects scams directly on Android devices, keeping private conversations local.
Google DeepMind and the Government Technology Agency of Singapore (GovTech Singapore) have announced a collaboration to develop an AI model that detects potential scams directly on Android devices. The initiative, unveiled during Singapore International Cyber Week 2026, aims to identify manipulative conversational patterns locally, without sending private message data to cloud servers.
The proposed system uses a fine-tuned version of Google’s lightweight Gemma model. According to Google DeepMind, the model is designed to analyze conversations on the device itself. This approach addresses the challenge of end-to-end encryption used by many messaging services, which prevents traditional cloud-based scanning of private chats.
When the model identifies suspicious interaction patterns, it displays a real-time warning on the user’s device. The warning urges users to pause before sharing sensitive information or transferring money. Google DeepMind states that the model accesses conversations only with user consent and that neither Google nor the Singapore Government can view user conversations through the deployment.
To manage battery efficiency, the system employs a two-stage structure. In the first stage, a lightweight classifier screens incoming messages. In the second stage, the fine-tuned Gemma model performs a contextual evaluation of a larger conversation snapshot to confirm a threat or discard a false alarm.
The model was fine-tuned using anonymized data from scam reports and intelligence feeds, excluding personally identifiable information. The training data included examples of local scams, such as impersonations of public institutions, banks, and local figures, as well as schemes circulating across the region. Google DeepMind says this helps the model distinguish ordinary conversations from the pressure and manipulation tactics used in social engineering.
The fine-tuning took place in a Google Cloud environment managed by GovTech Singapore. According to the announcement, the development process adhered to governance measures labeled "Sovereign Sandbox," "No Citizen Data," and "Stress-Testing." These measures ensured training data isolation and the exclusion of private citizen data.
The collaboration responds to significant financial losses from scams in Singapore. According to the Google DeepMind blog, scams have accounted for more than S$4 billion in losses since 2019. A mid-year report from the Singapore Police Force cited by Streamline Feed noted 16,821 scam cases from January to June 2026, with reported losses falling 17.9% to S$410.6 million compared to the previous year.
Police data indicates that in 80.8% of reported cases, victims made transfers themselves after being manipulated. The most common first contacts for scams in the first half of 2026 were messaging platforms (27.6%) and social media (26.7%).
Google DeepMind and GovTech Singapore have not announced a specific launch date, supported apps, or public accuracy results for the model. The company states that rollout within applications will be phased and developed in consultation with the Singapore government. Details on how users can install and use the model will be shared later.
Mark Chen, Director of Trust and Safety at GovTech Singapore, said: "Fighting scams shouldn’t mean sacrificing personal privacy. Working with Google DeepMind, we’re bringing advanced AI protection directly onto citizens’ devices. This allows us to catch highly targeted, local scam attempts in real time while keeping personal communications completely private."
This initiative represents a shift toward privacy-preserving AI security, moving detection from centralized servers to the user's device. By leveraging local processing, the model aims to protect end-to-end encrypted communications while providing real-time warnings against social engineering. However, the effectiveness of the tool remains unverified, as no independent accuracy metrics or deployment timelines have been released.
Sources
This article was drafted with AI assistance and checked against the sources above. Company claims are reported as claims. Cover image is AI-generated.
The day's checked news, by email
One short email a day with the top stories. Free, and you can unsubscribe in one click.
Or follow for the latest news:
No comments yet. Start the discussion.
Microsoft introduced new security features to monitor AI agents, warning that autonomous systems can pose insider risks even without being hacked. The company urges separating intelligence supply from control authority.
The Department of Homeland Security has ordered federal civilian agencies to remove DeepSeek AI products from IT systems and barred future procurements, citing cybersecurity risks.