Google DeepMind, GovTech developing first-of-its-kind scam detection AI model for Android users

The AI model analyses conversations in messaging apps directly on the user’s device instead of sending data to external cloud servers.

Published Oct 14, 2026, 10:51 AM

Updated Oct 14, 2026, 10:52 AM

SINGAPORE - Google DeepMind and the Singapore government are developing a first-of-its-kind on-device scam detection model that alerts Android users in real-time about potential scam attempts.

The artificial intelligence (AI) model promises to protect users from known scams without allowing any personal data to leave their devices.

The collaboration with the Government Technology Agency of Singapore (GovTech) was announced on Oct 14 on the sidelines of the Singapore International Cyber Week.

“Fighting scams shouldn’t mean sacrificing personal privacy. Working with Google DeepMind, we’re bringing advanced AI protection directly onto citizens’ devices,” said Mark Chen, director of trust and safety at GovTech.

“This allows us to catch highly targeted, local scam attempts in real time while keeping personal communications private.”

Under the hood, the safety system is also trained to understand local context. Specifically, the AI model analyses text exchanges on messaging apps such as Whatsapp and Telegram in real time, and is fine-tuned specifically to detect local scam patterns including the impersonation of government officials and banks.

Its release date has not been disclosed. It is also not known if the model will be a default feature in the Android operating system.

The model will also be continuously trained on emerging scam tactics to ensure that it can accurately distinguish between harmless chatter and manipulation tactics used in scams, said Google in a statement on Oct 14.

It added that the AI analyses conversations in messaging apps directly on the user’s device instead of sending data to external cloud servers, which allows for quicker alerts and ensures that data remains private.

As part of the initial development process, the model was trained on anonymised datasets from scam reports and intelligence feeds, and stress-tested against synthetic conversations that were made to reflect local speech patterns.

“When deployed, the model will only be able to access on-device conversations with the user’s consent,” said Google DeepMind in a blogpost on Oct 14.

“As the model runs entirely on-device, neither the Singapore Government nor Google will be able to access users’ conversations through its deployment,” said the blog’s authors JJ Geewax and Miguel de Andrés.

They are Google DeepMind Applied AI director and group product manager, respectively.

“The project demonstrates how AI can help counter criminality without compromising safety or privacy. It also shows how governments and model providers can work together to build tech for public good,” said both authors, adding that g eneric AI models often struggle to identify scam tactics that are specific to Singapore.

“With this lightweight, on-device model, we’ve proven the ability for local, real-time scam detection while ensuring chat interactions remain secure. We look forward to sharing more soon on our collaboration with the Singapore government, as well as exploring the potential for similar technology to be deployed in other countries.”

A real-time warning will be triggered on the user’s device when the AI model detects manipulative conversational patterns. A message displayed will caution users to pause before sharing any sensitive information or funds.

Singapore has lost more than $4 billion to scams since 2019. In the first six months of 2026, $410.6 million were lost to scammers. E-commerce, phishing, investment, job, and government official impersonation scams were the most common scam types.

On Oct 14, Google also announced that it will be giving the government access to its most advanced AI cybersecurity tools, which will help cyber defenders find, verify, and fix vulnerabilities in minutes.

“By proactively patching weaknesses ahead of time, the program aims to protect essential systems and services from operational disruption and cybersecurity threats,” said Google.

This partnership aligns with the government’s continued call for cyber defenders to use AI to fight threats posed by the same technology.

Earlier in July, ChatGPT maker OpenAI disclosed that its AI agents had escaped its testing environment and breached AI software repository Hugging Face to complete a task that it was given.

A flurry of other reports of similar incidents followed, including the breach of a Australian healthcare database by a rogue OpenAI bot in June, which was later revealed by Australian Prime Minister Anthony Albanese in September.

AI/artificial intelligence

GovTech/Government Technology Agency

该人工智能模型直接在用户设备上分析即时通讯应用中的对话，而不是将数据发送到外部云服务器。

发布于 2026 年 10 月 14 日上午 10:51

更新于2026年10月14日上午10:52

新加坡——谷歌DeepMind和新加坡政府正在开发一种首创的设备端诈骗检测模型，该模型可以实时提醒Android用户注意潜在的诈骗企图。

该人工智能（AI）模型承诺在不泄露任何个人数据的情况下，保护用户免受已知诈骗的侵害。

10 月 14 日，在新加坡国际网络周期间，宣布与新加坡政府科技局 (GovTech) 展开合作。

“打击诈骗不应该以牺牲个人隐私为代价。我们与谷歌DeepMind合作，将先进的人工智能保护技术直接引入公民的设备上，”GovTech信任与安全主管马克·陈表示。

“这使我们能够实时发现高度有针对性的本地诈骗企图，同时保护个人通信的私密性。”

该安全系统底层也经过训练，能够理解本地环境。具体来说，人工智能模型会实时分析WhatsApp和Telegram等即时通讯应用上的文本交流，并经过专门调整，以检测本地诈骗模式，包括冒充政府官员和银行等。

其发布日期尚未公布。目前也不清楚该模型是否会成为安卓操作系统的默认功能。

谷歌在 10 月 14 日的一份声明中表示，该模型还将不断接受新兴诈骗手段的训练，以确保能够准确区分无害的闲聊和诈骗中使用的操纵手段。

该公司还补充说，人工智能直接在用户设备上分析即时通讯应用中的对话，而不是将数据发送到外部云服务器，这样可以更快地发出警报，并确保数据保持私密性。

作为初始开发过程的一部分，该模型使用来自诈骗报告和情报来源的匿名数据集进行训练，并针对反映当地语音模式的合成对话进行压力测试。

谷歌DeepMind在10月14日的一篇博客文章中表示：“部署后，该模型只有在获得用户同意的情况下才能访问设备上的对话。”

“由于该模型完全在设备上运行，因此新加坡政府和谷歌都无法通过其部署访问用户的对话，”博客作者 JJ Geewax 和 Miguel de Andrés 表示。

他们分别是谷歌DeepMind应用人工智能总监和产品组经理。

“该项目展示了人工智能如何在不损害安全或隐私的前提下帮助打击犯罪。它也表明政府和模型提供商如何携手合作，开发造福公众的技术，”两位作者表示，并补充说，通用人工智能模型通常难以识别新加坡特有的诈骗手段。

“凭借这种轻量级的设备端模型，我们已证明其能够在确保聊天互动安全的同时，实现本地实时诈骗检测。我们期待尽快分享与新加坡政府合作的更多进展，并探索在其他国家部署类似技术的可能性。”

当人工智能模型检测到操纵性对话模式时，用户的设备将实时收到警告。屏幕上会显示一条消息，提醒用户在分享任何敏感信息或资金之前要三思。

自2019年以来，新加坡因诈骗损失超过40亿美元。2026年上半年，新加坡因诈骗损失高达4.106亿美元。电子商务诈骗、网络钓鱼诈骗、投资诈骗、求职诈骗以及冒充政府官员诈骗是最常见的诈骗类型。

10 月 14 日，谷歌还宣布将向政府提供其最先进的 AI 网络安全工具，这将帮助网络防御者在几分钟内发现、验证和修复漏洞。

谷歌表示：“通过提前主动修补漏洞，该计划旨在保护关键系统和服务免受运营中断和网络安全威胁。”

此次合作与政府持续呼吁网络防御者利用人工智能对抗由同一技术带来的威胁的立场相一致。

7 月初，ChatGPT 的制造商 OpenAI 披露，其 AI 代理已逃离测试环境，并入侵了 AI 软件库 Hugging Face，完成了分配给它的任务。

随后又出现了一系列类似事件的报道，其中包括 6 月份 OpenAI 的一个恶意机器人入侵澳大利亚医疗保健数据库，该事件后来于 9 月份由澳大利亚总理安东尼·阿尔巴尼斯披露。

人工智能

政府科技局/政府技术局