Google DeepMind and Singapore Unveil On-Device AI Scam Detection Tool

Google DeepMind and the Government Technology Agency of Singapore (GovTech) have jointly developed an on-device scam detection model designed to protect mobile users from fraud without compromising their privacy. The collaboration, announced in a Google DeepMind blog post by JJ Geewax and Miguel de Andrés and presented at Singapore International Cyber Week (SICW 2026), was carried out as part of the National AI Partnership between Google and the Singapore Government.

Scams Dominate Singapore's Crime Landscape

Scams make up over half of Singapore's reported crimes and have resulted in more than S$4 billion in losses since 2019. Scam tactics have also shifted from simple phishing to complex social engineering, in which attackers manipulate victims through convincing conversations before requesting money. Such conversational scams now account for up to 80% of all losses. Because these schemes unfold through ordinary back-and-forth chat rather than a single suspicious link, they are harder for conventional message filters to catch.
To address the threat, the two organizations optimized the model to run on typical smartphones in Singapore, tackling an urgent safety challenge while preserving citizens' privacy.

Local Training Data, On-Device Analysis

Generic AI models often struggle with localized deception techniques, making scam detection harder. To close this gap, Google DeepMind's lightweight Gemma model was fine-tuned on local scam conversations, ranging from impersonations of public institutions, banks, and local figures to trending regional schemes. Gemma is Google DeepMind's family of lightweight open models built to run on everyday hardware, a profile that suits smartphone deployment. This training enables the model to distinguish benign social chatter from genuine manipulation.

The model also processes conversation data directly on the device. Traditional threat detection relies on server-side scanning of messages in the cloud, but many messaging platforms are end-to-end encrypted, meaning providers cannot inspect private conversations. In practice, that has left scam conversations inside encrypted apps invisible to server-side defenses. By moving threat analysis onto the device itself, the system can analyze conversation patterns faster, detect manipulative intent, and allow citizens to act before financial or credential exposure occurs.

"Fighting scams shouldn't mean sacrificing personal privacy," the announcement stated. "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."

How the System Works

When the model detects manipulative conversational patterns, it triggers a real-time warning directly on the user's device, advising them to pause and exercise caution before sharing sensitive information or funds.

Running a complex scam classifier on modern smartphones is a significant engineering challenge, with battery efficiency a critical concern. The system therefore relies on a two-stage pipeline:

• Stage 1: A lightweight, efficient classifier screens inbound messages using minimal power, focused on catching potential threats.
• Stage 2: Contextual evaluation with the fine-tuned Gemma model, which analyzes a larger snapshot of the conversation to confirm a threat or discard a false alarm.

Built on Strict Governance

The collaboration was presented at SICW 2026 as a model for messaging platforms balancing user protection with end-to-end encryption, and strict governance was maintained throughout development:

• Sovereign Sandbox: Fine-tuning took place within GovTech's managed and maintained Google Cloud Platform (GCP) environment, keeping training data isolated.
• No Citizen Data: Training strictly excluded private or personally identifiable information, using anonymised datasets derived from scam reports and intelligence feeds.
• Stress-Testing: The model was tested against synthetic dialogues and local corpora to cover diverse scam categories and evolving impersonation tactics.

The Future of On-Device Safety

According to the blog post, the project demonstrates how AI can help messaging services counter criminality, and shows that neither safety nor privacy needs to be compromised when protective AI runs directly on-device. It also illustrates how governments and model providers can work together to build technology for the public good.

Messaging platforms and operating system vendors can directly integrate these lightweight models to safeguard users locally, keeping chat interactions secure without exposing private data. For now, the blog post frames the project as a demonstration of what on-device protection can achieve; whether the industry moves from that possibility to shipped integrations will be the next development to watch.

Source: Google DeepMind Blog — On-Device AI Against Scams: Google DeepMind & MDDI