Select Language:
Several popular apps, including Tencent’s WeChat and banking apps from the Agricultural Bank of China and China Construction Bank, have experienced disruptions after an AI-powered assistant created by a major tech company developed by ByteDance attempted to operate them automatically.
Numerous users with the new Nubia M153 smartphone reported being forcibly signed out of these apps after the mobile assistant tried to access WeChat, triggering a warning message that read “login environment abnormal.” Additionally, the banking applications from AgBank and China Construction Bank displayed alerts advising users to disable the assistant before they could proceed.
The Nubia M153 is the debut model from ZTE, a Chinese company specializing in telecommunications equipment and consumer electronics. It features the Doubao Mobile Assistant and was launched on December 1 at a price of 3,499 yuan (approximately $495). Currently, the device is only available to developers and industry professionals.
In response to the issues, WeChat clarified that it did not take any specific actions, indicating that the app’s security protocols might have triggered existing risk controls due to the assistant’s activity.
The Doubao Mobile Assistant’s key feature is its ability to switch smoothly between various applications, handle reservations, place online orders, download multiple files simultaneously, monitor package deliveries across different platforms, and respond to messages—all through simple voice commands from the user.
Industry experts pointed out that the Nubia M153 granted the assistant a high-level permission called INJECT_EVENTS, a privilege usually reserved for system components designed by phone manufacturers. This permission allows the assistant to read on-screen content, simulate taps and clicks, and perform other automatic interactions.
One technician explained that the assistant is integrated into the device’s operating system rather than functioning as a third-party app. The company behind the assistant emphasized that there is no hacking involved; the permission is a system-level feature that users must explicitly authorize for the assistant to operate. They also confirmed that the tool does not engage in sensitive activities like payments or identity verification and does not upload screen data for training purposes.
The core of the ongoing AI competition involves gaining access to private user data, with device manufacturers, app developers, and traditional software companies vying for control. Internet firms view proprietary data as fundamental to their operations and are cautious about third-party apps obtaining high-level permissions.
Recently, a standards committee published guidelines stating that AI agents should not bypass security checks of third-party apps by simulating user interactions. Developing strong dual-authorization systems remains an ongoing process.
In the past month, a major online retailer filed a lawsuit against an AI startup claiming that its AI browser tool violated terms of service by shopping on behalf of Google users. The startup responded by criticizing the lawsuit as excessive and emphasized that users should be free to utilize AI tools independently.
ChatGPT
Perplexity
Gemini AI
Grok AI
