Google and Singapore Team Up on Cyber Defence and Scam Detection
Google and the Singapore government widened their joint work on digital safety at Singapore International Cyber Week, announcing two efforts that draw local agencies and the company's newest security models closer together. Singapore will join Google's Fairwind programme for cyber defenders, and Google DeepMind will work with the Government Technology Agency on a scam detection model that runs on a person's own phone.
Under Fairwind, defenders get access to a model called Gemini 4 Argon and to a harness known as CodeMender. Google says the combination lets security teams find, verify and fix vulnerabilities at agentic scale, so work that once took weeks of manual effort becomes a set of verified patches generated in minutes inside an organisation's own cloud environment. The point is to close weaknesses before they can disrupt essential services.
And the programme is aimed at professional defenders rather than at consumers, since the systems it protects sit behind government services and critical infrastructure. Singapore's participation gives its agencies early use of tooling that Google describes as its most advanced cybersecurity offering.
The announcement follows earlier work between the two. Google refreshed a memorandum of understanding with the Cyber Security Agency, and the pair ran an AI Agents Sandbox described as a global first, set up to study how computer use agents behave in real settings.
Google also pointed to its consumer defences, where Google Play Protect has blocked more than 3.3 million installation attempts of potentially harmful apps across more than 770,000 devices, and stopped 68,000 apps from being misused for financial fraud. Its Be Scam Ready game, which trains players to spot manipulation, was credited with lowering players' overall scam risk by about 20% after ten minutes of play, with those players 47% less likely to mistake a scam for a legitimate message and 24% less likely to flag a genuine one wrongly. A second version halves the play time.
But the more unusual of the two projects puts the typing on the handset, with Google DeepMind and GovTech building a model for Android phones that reads conversations locally and warns the user while a scam is playing out. It uses lightweight Gemma models trained on scam stories from Singapore and the region, from impostors posing as banks and public institutions to schemes that are trending at the time. Because the analysis stays on the phone rather than going to a server, private chats are not exposed.
Two of the people involved framed that as a privacy question. "Fighting scams shouldn't mean sacrificing personal privacy," said Mark Chen, director of trust and safety at GovTech Singapore.
What users will not find yet is a way to get it. Google has not said when the model reaches phones, whether it arrives in a system update or as something people install, or whether Singapore gets it first. The company says only that more detail on how to install and use it will follow.
So the announcement covers two fronts, hardened tools for professional defenders and a lightweight model for everyone else. The scam model is the piece worth watching, because it moves detection onto hardware people already own and keeps their conversations there. When either effort reaches ordinary users is the question the announcement leaves open.
Google announced two Singapore collaborations at Singapore International Cyber Week, the Fairwind programme and a scam detection model built with GovTech
Fairwind gives defenders the Gemini 4 Argon model and the CodeMender harness, aimed at generating verified patches in minutes rather than weeks
The scam model uses lightweight Gemma models tuned on local scam narratives and runs on the phone, so private chats are not sent to servers
Google Play Protect has blocked more than 3.3 million installation attempts across more than 770,000 devices, and Google has not said when the scam model will reach users
Source: Google
