{
  "id": 14376558,
  "title": "Singapore wants Google’s AI to catch scammers before victims hit send",
  "url": "https://urgent.news/2026/10/14/singapore-wants-googles-ai-to-catch-scammers-before-victims-hit-send",
  "topic": "ai",
  "section": "AI",
  "published": "2026-10-14T05:00:10.000Z",
  "source": {
    "name": "e27",
    "slug": "e27",
    "url": "https://e27.co/singapore-wants-googles-ai-to-catch-scammers-before-victims-hit-send-20261014/"
  },
  "original_language": "en",
  "account": "Singapore's efforts to combat scams have typically focused on external measures such as police operations, bank measures, and rules dictating platform blocking. However, the city-state is now seeking to incorporate a solution within the phone itself. On the sidelines of Singapore International Cyber Week (SICW) 2026, Google announced two initiatives in partnership with the Singapore Government. The first invites Singapore to join Google's Fairwind Programme, a selective scheme providing vetted defenders with Google's most advanced cyber models, including its new Gemini 4 Argon model. The second collaboration pairs GovTech Singapore with Google DeepMind to develop an on-device AI model for Android that identifies scam attempts in real-time without transmitting private conversations to the cloud. Both initiatives are positioned as defensive measures, but they also increase Singapore's reliance on a single US tech company at a time when access to frontier AI is becoming a geopolitical tool. The Fairwind Programme, launched in September, began with limited access to governments and trusted partners, starting with Gemini 3.8 Flash Cyber integrated with Google's CodeMender agent. Since then, Gemini 4 Argon has become the program's main attraction. Fairwind partners and internal Google teams receive a version of Argon without cyber guardrails, which normally block requests resembling hacking. Organizations can only grant Argon access to their internal cybersecurity, incident response, or penetration testing teams and must monitor employee access and usage. Google emphasizes speed, claiming that with its model, vulnerabilities can be fixed minutes rather than weeks. However, this is a vendor claim, not a verified outcome. Analysts have pointed out that the most tangible evidence would be merged-patch or time-to-remediation results from Fairwind partners, which Google has not published. Additionally, there are concerns about who can view the model's actions. Google asserts that it monitors Argon's chain of thought and actions, stopping execution when necessary, but this implies a foreign company has some insight into how a government's defenders examine their systems. Google confirms that Argon operates on a zero data retention policy when used as a managed model on Gemini Enterprise, but the specifics of Singapore's participation were not disclosed. While Google is not the only lab limiting access to its most advanced cyber tools, other entities like Anthropic and OpenAI have also launched similar programs. These initiatives involve restricting access to governments and companies, creating access itself as the product. This access can also be withdrawn at any time. The scam problem is primarily a human issue. Although Singapore's scam-related losses decreased by 19% in 2025 to S$913.1 million (around US$715 million), the composition of these losses reveals why the government is considering device-based solutions. In 81.8% of 2025 scam cases, victims transferred the money themselves instead of having their accounts taken over. Most losses are prevented through takedowns and bank freezes, which are often too late. The decisive moment is when a manipulated person presses \"send.\" To address this, GovTech-DeepMind has developed a lightweight Gemma model, tailored to local scam narratives like impersonation of public institutions, banks, and local figures. This model runs entirely on the handset and issues a warning on screen when a conversation exhibits high-pressure social engineering tactics. According to Mark Chen, Director of Trust and Safety at GovTech Singapore, the approach maintains personal privacy. However, important questions remain unanswered, such as the model's launch date, accuracy, false-positive data, and how it reads conversations within end-to-end encrypted apps like WhatsApp or Telegram, which inherently hide content from anyone except the sender and recipient. The model is currently Android-only in a market dominated by iPhones, leaving a significant portion of users unprotected. This timing is not coincidental. Singapore has recently tightened scam rules for messaging, social media, and e-commerce platforms, with Google Messages and Google Meet among the designated services required to comply by January 31, 2027, or face fines up to S$10 million (US$7.83 million). Google's on-device shield serves as both Google's compliance story and its public-safety initiative. It also enters a crowded field, as Singapore's ScamShield app and caller-ID players such as Taiwan's Gogolook (Whoscall) and Truecaller have already gained significant popularity in the region. Google's advantage lies in its distribution, as a model integrated into Android reaches users who might otherwise not be covered.",
  "summary": "Singapore’s war on scams has mostly been fought far from the victim: in police operations centres, bank back offices and, more recently, in rulebooks telling platforms what they must block. Now the city-state wants part of that fight to happen inside the phone itself. Google on Wednesday announced two initiatives with the Singapore Government, unveiled […] The post Singapore wants Google’s AI to…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}