By 2026, China’s consumer-grade AI agents are poised to enter a new era. While the underlying technology has reached maturity, the management of cross-App permissions has emerged as a critical factor for further development. Alipay’s Abao, WeChat’s Xiaowei, and Doubao II (from Nubia) have each carved out distinct paths in this evolving landscape:
Alipay’s Abao implements a “dual-track system” with “one-click switching” functionality. By default, it prioritizes user instructions by enabling authorized service calls, and only sensitive operations necessitate separate user confirmation.
WeChat’s Xiaowei, in contrast, places a strong emphasis on “permissions” from the outset. It establishes a three-tier boundary framework and adheres to the guiding principle of “assisting but not taking over,” ensuring users maintain control.
Nubia’s Doubao II introduces the Screen Automation Execution Protocol (SAEP), striking a balance between automation and user control. It limits the scope of automated operations through a 30-day public notice period and incorporates a flexible exit mechanism that allows developers to opt out if needed.
Collectively, these three AI agents aim to establish reasonable boundaries for AI agent behavior. Their approaches reflect a careful balance between fostering technological advancement and mitigating potential risks in a rapidly evolving digital ecosystem.