- On-device AI agents with system-level permissions can trigger third-party app security mechanisms even without malicious intent
- ByteDance's Doubao denied any unauthorized clicking, cheating or simulated operations in Tencent's Honor of Kings
- The incident underscores the lack of standardized trust frameworks between AI agent platforms and major app ecosystems in China
Anti-Cheat Conflict Emerges on New AI Phone
Users of ZTE's Nubia NaviX Ultra, the first mass-produced smartphone built around an on-device AI agent, reported being forced offline while playing Tencent's Honor of Kings after the game flagged an 'abnormal device environment.' The phone ships with ByteDance's Doubao Phone Assistant, which uses an on-device large model to perform cross-app autonomous operations.
The incident drew attention because the NaviX Ultra had only recently launched, with pre-orders beginning September 14 and exceeding 370,000 reservations on JD.com alone before its official release on September 16.
Doubao Denies Misconduct, Seeks Resolution With Tencent
Doubao Phone Assistant responded in its official community with an apology, stating that the system did not intervene in Tencent's game at any point and that the AI did not engage in unauthorized clicking, cheating plug-ins, or simulated operations. The team said it is continuing to communicate with Tencent and advised affected users to pause login attempts or file account appeals through official channels.
Trust Gap for On-Device AI Agents
The episode illustrates a structural challenge for China's push toward on-device AI agents: system-level assistants that schedule tasks across multiple applications can resemble automation tools that anti-cheat systems are designed to block. Establishing trust boundaries between AI agent platforms and third-party app security frameworks has become a prerequisite for scaling this technology commercially.
Sources & context
Go to the original material. Company claims remain attributed to their sources.
01
