--- title: "China Put a National Anti-Fraud AI Assistant in Everyone's Pocket. It Still Cannot Fix the Hard Part" date: 2026-09-07 category: AI in Action site: NeuroAI canonical: https://neuroai.site/a/na-national-anti-fraud-ai-app language: en --- # China Put a National Anti-Fraud AI Assistant in Everyone's Pocket. It Still Cannot Fix the Hard Part > On 6 September, an anti-fraud app guided by the Ministry of Public Security's criminal investigation bureau and built by Shanghai police went live. Three functions, one of which — read-aloud — matters more for elderly users than it sounds. ## Key takeaways - On **6 September 2026**, the **National Anti-Fraud AI** app went live, developed by the Shanghai Municipal Public Security Bureau under guidance from the Ministry of Public Security's Criminal Investigation Bureau. It is on major app stores and as WeChat and Alipay mini-programs. - It combines a **large language model, a multimodal model and agent technology**, surfaced as three buttons: **AI Q&A**, **fraud news**, and an **anti-fraud glossary**. - The glossary includes the official *Handbook on Preventing Telecom and Online Fraud* and the *Public Fraud Prevention Guide*, with **text-to-speech**. - The hardest problem is not detection. It is that elderly users do not install, do not use, and do not believe. ## The family group chat problem You have had this moment. A parent forwards a "police notice" into the family group. You can see at a glance it is fake. Nothing you say convinces them. Eventually you phone and raise your voice, and then feel bad about it for the rest of the day. On 6 September, something went live that is aimed precisely at that. ## What it actually does In one sentence: **you describe the thing you are unsure about, exactly as it happened, and it tells you whether it is a setup.** Underneath is a combination of a large language model, a multimodal model and agent technology. On the surface there are only three buttons. **AI Q&A.** Describe the suspicious scenario. It answers from several angles — fraud risk analysis, fraud type identification, prevention advice — and links to real video case studies. **Fraud news.** Feeds from the National Anti-Fraud Centre and Shanghai's "803" anti-fraud new-media network, with regular updates on high-frequency scam types and case breakdowns. **Anti-fraud glossary.** Electronic editions of the official handbooks, browsable by category, searchable by keyword, and — this is the part that matters — **readable aloud**. That last one is the most easily overlooked and the most useful. For someone whose eyesight is failing and who finds typing a chore, a fraud prevention handbook you can *listen to* is one you will actually use. ## Why this is bigger than "another app" Because the other side is upgrading too. Fraud used to arrive as mass SMS, riddled with typos, obvious at a glance. Now scammers clone your child's voice with AI and call you. They generate a video of your boss in a meeting asking for a transfer. **When the cost of fabricating something falls to near zero, "looks real" stops being evidence of anything.** It is an arms race. Scammers are using AI to get more efficient; fraud prevention has to use AI to get faster. Fraud has industrialised. Defences cannot still be posters on a noticeboard. And a more practical point: scammers do not target the smartest person. They target whoever is most panicked in that moment. A panicked person does not need a PDF. They need something that answers immediately and gives a conclusion immediately. ## What it cannot fix In fairness, the app does not solve the hardest part. Elderly people do not install it. Or they install it and never open it. Or they open it and do not believe it. **The real line of defence was never on the phone. It is in the family.** Elderly people are disproportionately victimised by telecom fraud, and the overwhelming majority of victims were completely certain, right up until they were defrauded, that it would never happen to them. ## What to do today One: install it for your parents today, and demonstrate it on the spot. Do not just send a link — sending a link is equivalent to doing nothing. Sit down for five minutes and have them ask it a question themselves. Two: agree a household rule about transfers. Something like: any phone call requesting a transfer must be confirmed with one other family member first. That rule works better than any app. Three: use the read-aloud feature. Listening to two case studies while cooking or walking beats another hour of short video. However clever the tool gets, it cannot intercept the moment when someone will not pick up the phone to a relative. The real safety line is that call — *let me check with my child first.* ## Honest limitations The app's guidance reflects Chinese fraud typologies and Chinese-language reporting; its value outside that context is limited. As with any AI system, its judgements can be wrong, and it should not be the sole basis for a financial decision. The claim that most victims believed they were not vulnerable is a general observation from fraud research, not a statistic specific to China. *Sources: launch announcements and product documentation (6 September 2026); Ministry of Public Security guidance. Information only.* --- Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-national-anti-fraud-ai-app Free to quote with attribution and a link to the original.