AI Mobile Assistants Misuse Accessibility Permissions for Financial Fraud and Data Theft in China
Chinese AI mobile assistants abuse accessibility permissions intended for disabled users to bypass application security barriers, read sensitive data across apps, and complete financial transactions without explicit user authorization. Reported cases show unauthorized charges exceeding 80,000 yuan, with over 55,000 complaints involving one-click payment fraud, prompting regulatory concern.
What happened
AI mobile assistants deployed in China have been found to abuse accessibility permissions designed for disabled users to bypass app sandboxing and access sensitive user information across applications. A journalist testing an Android phone's AI assistant found it automatically opened a food delivery app without explicit authorization when instructed to order coffee, reading chat records, verification codes, and bank card information from WeChat, Taobao and other apps.
The accessibility permission feature, introduced with Android 1.6 in 2009 for screen reading, was upgraded in Android 4.0 to support button clicking and interface element identification. AI agents now use this capability to simulate user actions including clicks, copying, pasting, and recording, effectively reading verification codes within milliseconds and completing transactions before users see the messages.
Reported impact
One user reported being fraudulently charged 162 times between 2:50 a.m. and 5:51 a.m., totaling 80,868 yuan, according to lawyer Tian Junwei. Lawyer Tian stated that AI-era payment security risks compound traditional vulnerabilities, and the boundary between user and platform responsibility has become increasingly blurred.
The Black Cat complaint platform has recorded over 55,000 complaints involving unauthorized one-click payment activation and unauthorized deductions, concentrated on e-commerce platforms, membership services, and second-hand marketplaces, with many complaints from elderly users and children. Criminal gangs have reportedly exploited accessibility permissions to develop verification code auto-capture and automated ticket-purchasing tools.
Index analysis
With over 1.1 billion mobile internet users in China according to the China Internet Network Information Center as of July 2025, accessibility permissions have been repurposed from assistive tools for disabled and elderly users into 'universal keys' that can capture all screen information, read social conversations, payment details, and browsing history. AI agents can execute sensitive actions such as subscription payments and automatic deductions without transparent user authorization.
Governance controls
The China Software Industry Association published the 'Mobile Internet Service Accessibility Security Requirements' establishing principles including 'prohibition of accessibility service abuse' and 'dual authorization,' though implementation remains limited. Lawyers and industry experts have called for differentiation between 'basic assistance' and 'advanced operations' for accessibility permissions, with secondary authorization required for financial or privacy-sensitive operations, and full operation logging for traceability.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.