AI Smartphones: Security Over Intelligence

The true battleground in the AI smartphone market is shifting beyond advanced AI intelligence competition to building a reliable architecture against unprecedented security threats posed by AI agents.

AI smartphones are evolving beyond simple app execution to center around AI agents. AI agents understand user intent and perform tasks like making reservations, purchases, and sending messages on their behalf, but this convenience is accompanied by new security threats. The core issue lies in AI agents operating as de facto 'superusers' within the device, as their structure relies on the highest operating system privileges, such as reading screens, controlling interfaces, and moving data between apps.

Security threats in the AI agent era differ in nature from traditional software bugs. While conventional attacks breach code directly, attacks targeting AI agents focus on 'manipulating intent.' A prime example is command hijacking (prompt injection).

Amid security concerns, major platforms like Alipay and WeChat are hesitant to grant system access privileges to AI agents. Meanwhile, Google is pursuing a structured architecture that limits AI to performing only defined functions through standardized interfaces (App Functions), rather than allowing it to observe and click everything.

Vivian Toh, Editor-in-Chief at Chinese tech news startup TechTechChina, diagnosed that the ultimate winner of the AI smartphone race will not be the smartest AI, but the company that builds the most trustworthy security architecture. Toh emphasized, 'The true winner of this competition will be the one who can convince all app platforms, regulatory bodies, and users worldwide that they are worthy of entrusting the keys to devices containing their bank accounts and personal information.' Toh added that an architecture featuring minimal privilege design, clear audit trails, and strict user controls for high-risk actions would be the prerequisite.