A smartphone with an AI agent baked into the operating system sounds like the future. In China, it lasted about a week before the country's biggest apps shut it out.
In December 2025, ByteDance and ZTE launched the Doubao phone, a device built around an AI agent that lives at the system level. The agent has a permission called INJECT_EVENTS, which gives it blanket access to read screens, interact with apps, and act on behalf of the user across the entire device. Think of it as giving one piece of software a master key to everything on your phone.
The Backlash Was Immediate
Within days, WeChat, Taobao, Alipay, and multiple banking apps blocked the device entirely. Videos circulated showing users' bank balances visible across synced devices, which turned public skepticism into outright alarm. Banks pointed out a fundamental problem: they cannot distinguish between actions taken by the user and actions taken by the AI agent. From a fraud prevention standpoint, that is a serious gap.
Critics described the agent's permissions as giving a "burglar" the keys to your digital life. The concern is not hypothetical. If a system-level agent can see and interact with banking apps, payment platforms, and messaging services, the attack surface for fraud and credential theft expands dramatically.
A Competition Problem Disguised as a Privacy Problem?
Defenders of the Doubao phone argue the backlash is less about genuine security and more about competitive territory. China's digital economy runs through superapps like WeChat and Alipay, and those platforms have little incentive to let an outside AI agent sit between them and their users. If ByteDance's agent can operate inside WeChat, it threatens Tencent's control over user attention and data. The privacy argument, in this view, is a convenient shield for competitive self-interest.
There is probably truth on both sides. The security risks are real, but so is the commercial motivation to block a rival's technology.
The Regulation Question
Chinese scholars and industry experts are now proposing guardrails: mandatory suspension of the agent during sensitive transactions like banking, on-device processing for private data instead of cloud transmission, stronger consent mechanisms, and antitrust oversight to prevent dominant platforms from using security as a pretext to block competitors.
This matters well beyond China. The same tension will play out everywhere as AI agents get deeper system access. Apple's Siri, Google's Gemini integration, and OpenAI's partnerships with device makers will all face the same question: how much access should an AI agent have, and who decides when it is too much?
The Doubao phone is an early, messy test case. The answers China arrives at will likely influence how regulators in Europe, the US, and elsewhere think about agentic AI on personal devices.