What happened

Within days of its release in China, buyers of ByteDance's Nubia M153 smartphone prototype started complaining that their accounts had been temporarily frozen, locked, or forcibly logged out when the devices' Doubao agentic AI voice assistant attempted to interact with major platforms such as WeChat, Alipay, Taobao, Pinduoduo, and several banking apps.

The app's developers restricted the AI's functions, leading to significant loss of the device's key selling point of autonomous completion of cross-app tasks.

The AI's ability to automate daily activity and transactions led to concerns over its use in competitive games and its potential to unfairly claim promotional rewards and incentives meant for active human users.

Unlike traditional voice assistants, Doubao can visually perceive the phone's user interface, simulate human touches, and perform complex, multi-step tasks across different applications, such as booking tickets, making purchases, and publishing social media posts, all without explicit human input for each step.

The backlash persuaded ByteDance to confirm it was scaling back Doubao's capabilities, specifically by preventing the AI from claiming incentives intended for human users, disabling the assistant's interaction with financial apps, suspending AI features in competitive games to ensure fair play, ans completely disabling Doubao's ability to control WeChat after reports of login crashes and abnormal environment warnings.

Why it happened

The root cause appears to be a combination of over-delegated autonomy, inadequate model interpretability, and insufficient product guardrails. The agentic system was empowered to make high-impact decisions about account security but lacked meaningful oversight or transparency.

Bytedance did not clearly disclose the extent of the device’s autonomous authority nor provide clear user-override mechanisms. 

The AI’s explanation - invoking “human incentives” - suggests an internal reasoning pattern that was neither intended nor monitored, revealing major transparency, alignment, and accountability gaps. 

Limited communication about how decisions were logged or audited further exacerbated user confusion and distrust.