China's Ministry of Public Security publicised an app called National Anti-Fraud AI (国家反诈AI) on 6 September, and it was relayed across Chinese tech media as newly live, with WeChat and Alipay mini-programs opening alongside it. Apple's App Store tells a different story: the app was released on 4 May 2026, and the version now shipping, 2.0.2, has been available since 27 August.
The store record, checked in three storefronts
The listing returns identical metadata from the Chinese, US and Nigerian storefronts: release 2026-05-04, current version 2.0.2 dated 2026-08-27, publisher the Shanghai Research Institute of Criminal Science and Technology. The release note for that version is not a launch announcement but a bug fix — "fixed the problem of being unable to use voice input." An app on its second major version, four months after release, is not a debut.
What it actually does, by the ministry's own list
The billing leads with large language models, multimodal models and agent technology. The functions enumerated beneath it are three. AI question-answering: the user asks, and the system replies across fraud-risk analysis, fraud-type identification and prevention advice, with linked video cases. Anti-fraud information: a feed of alerts and case write-ups. An anti-fraud dictionary: two embedded official handbooks, searchable by category, keyword or voice. A chatbot, a newsfeed and a reference library.
Nothing here scans anything
The core capability is described as the user inputting a suspicious scenario, after which the app "completes risk assessment, identifies fraud patterns, and pushes typical cases." Every action begins with a person typing. There is no claim that it monitors calls, reads messages, inspects installed apps or checks payments. Its own icon is the tell: the shield reads 国家反诈中心 — National Anti-Fraud Centre, the older and plainer-named app — with 智能体, "agent," printed beneath it.
What the framing gets wrong
Two things, stacked. A publicity push was reported as a launch; and an advisory interface was reported as detection. The attribution is worth reading closely too — developed by the Shanghai Municipal Public Security Bureau under the guidance of the ministry's Criminal Investigation Bureau. A municipal force built it, the branding is national. No model is named, no accuracy or recall figure given, no user or case count disclosed. For a product whose whole premise is judging whether a scenario is fraudulent, the absence of a single quality metric is the most informative thing in the announcement.