Singapore Built an AI App to Catch Scams Before You Pay

Those transfers weren't forced. They were requested, and the requests were convincing enough that people complied. That pattern, a fraudulent message followed by a voluntary payment, is the gap Singapore's Government Technology Agency set out to close.

GovTech launched AntiScam Pango in late 2026, a free Android beta app built with Google DeepMind that watches conversations across WhatsApp, Telegram, and SMS notifications in real time. The analysis runs on the device itself, so messages never leave the phone. It also doesn't need a sender's number to appear on any blacklist. It looks at the content and shape of a conversation, flagging pressure to pay, impersonation language, and other patterns before a transfer can go through.

Singapore Police Force data for the first half of 2026 shows that victims initiated their own payments in 80.8% of messaging scam cases. The scammer never had to touch a bank account. The message was enough.

What the App Does

AntiScam Pango runs in the background, reading notification content rather than opening apps or storing chat logs. When it detects a suspicious pattern, it alerts the user before they act. The goal is to interrupt the moment between reading a message and responding to it, which is exactly when most losses happen. Keeping the analysis on-device also means sensitive conversation data stays private, a meaningful design choice given that the app monitors personal messaging platforms by necessity.

There is one thing AntiScam Pango cannot do, and it matters. The app can flag a suspicious message from a stranger. It cannot tell you whether the person claiming to be your friend, your sibling, or your colleague actually is. When a scammer takes over a contact's account or copies their name and photo, the message may look entirely normal. No blacklist catches that, and no AI pattern-match on the text alone resolves it.

That gap is where a simple identity check fills in. If someone messages you asking for money or personal information and claims to be someone in your trusted circle, ask them for the words. Trust Onion gives people in a shared group three private rotating codewords, and a voice clone or account hijacker cannot produce words they never had access to.

What to Do When a Message Feels Wrong

An app alert is a prompt to pause, not a verdict. If a message asks you to pay, move money, or share account details, treat it as unverified regardless of the sender's name. For a contact you know personally, the Trust Onion three-word check is a direct way to confirm it's really them. For a bank, delivery company, or government agency, hang up or close the chat and call back using a number from the official website or the back of your card. Even a successful callback to a named organization doesn't prove the original message was legitimate, so always use a number you've found independently.

Singapore built AntiScam Pango because the problem is large enough to require infrastructure, not just advice. But the statistic underneath the launch is the one worth remembering: in most cases, no one forced anyone to pay. The message just had to be believable long enough.

A controlled voice-cloning safety demonstration

Try our tool to see just how easy voice cloning is. This controlled safety demonstration exists only to show how easy voice cloning is and to make the risk tangible. It is not a real incident or proof of anyone's identity. Your submitted sample is handled transiently to generate the fixed demonstration audio. Trust Onion does not identify you or the person from the sample, and does not keep the submitted recording or generated audio. The temporary voice clone is deleted from our voice provider as soon as the demo finishes; if that deletion fails, it is retried automatically until the provider confirms it.