Agentic AI: Smartphone Privacy Risks

Avatar
Lisa Ernst · 07.12.2025 · Technology · 5 min

The development of AI assistants on smartphones reaches a new stage with the Doubao agent on the Nubia M153. For the first time, a major provider is attempting to bridge the gap from assisting functions to actively completing tasks directly in the operating system. This represents not only a technical preview but also an early model for a new interaction logic: voice input leads to cross-app actions.

Introduction

The ongoing integration of Artificial Intelligence into mobile devices is transforming the interaction between users and smartphones. While previous assistants primarily played supporting roles, the introduction of agentic AI hints at a more profound transformation. The ByteDance Doubao AI phone und das Nubia M153 are pioneering in this regard. They signal the transition from passive assistance to active, system-wide task execution. This is less a single gadget and more a model for a new interaction logic: voice in, cross-app action out, as reported by South China Morning Post .

The Nubia M153, a "engineering prototype", , is currently available in limited quantities only. Its price of 3,499 yuan (around $495) positions it in the premium mid-range device segment, according to Reuters.

Doubao AI Assistant

The core component of the Nubia M153 is the Doubao Mobile Assistant. . This is integrated at the OS level with the device and can perform tasks via voice. These include finding content, booking tickets, or controlling apps, as reported by Reuters . Doubao can also edit photos, compare prices across shopping apps, book restaurants, and, with user consent, initiate purchases, as highlighted by South China Morning Post .

ByteDance has clarified that the company does not intend to build its own smartphones. Instead, agent functionalities are to be distributed through partnerships with manufacturers, according to Reuters.

Security Concerns

Cross-sectional – The security of AI systems is a central concern in the development of agentic AI smartphones.

Source: scworld.com

The security of AI systems is a central concern in the development of agentic AI smartphones.

The technical integration of the Doubao assistant is exciting but also delicate, especially its access to system-level permissions, as notes Yicai Global . Industry experts point out that the assistant on the Nubia M153 uses the Android permission INJECT_EVENTS. This is usually reserved for system components and allows for simulating input and reading screen content, according to Yicai Global.

ByteDance has responded, stating that this system-level capability is only used with explicit consent. The company assures that no malicious hacks are involved and that sensitive operations such as payments or identity checks are not performed on behalf of users. Furthermore, the team emphasizes that screen content is not stored in the cloud and is not used for training, as reported by Yicai Global .

Cross-sectional – Privacy challenges in agentic AI include consent management and data minimization.

Source: xenonstack.com

Privacy challenges in agentic AI include consent management and data minimization.

This is precisely where the "agentic AI smartphone privacy concerns", , familiar from the desktop and enterprise context, arise. The deeper an agent can access system and apps, the more important clear boundaries, logging, and verifiable consent become, according to Privacy International.

Platform Reactions

The initial reactions from major platforms show how explosive OS-level automation can be in app ecosystems, as analyzed by South China Morning Post . User reports indicate that WeChat and some major banking apps displayed warnings or made logins more difficult when Doubao functions were active, according to Yicai Global. WeChat stated that it had not taken specific action against Doubao; it is possible that internal security mechanisms were triggered, according to Yicai Global.

ByteDance, in turn, announced that it would scale back the assistant's capabilities, particularly deactivating interactions with financial apps and pausing AI functions in competitive games to address security and fairness issues, as reported by South China Morning Post . Additionally, ByteDance intends to prevent the device from automatically claiming incentives meant for human users. This suggests that agents can alter not only security but also market and incentive systems, according to South China Morning Post.

The Future of Mobile AI

Cross-sectional – The market for AI-enabled smartphones is expected to grow significantly in the coming years.

Source: fr.techtribune.net

The market for AI-enabled smartphones is expected to grow significantly in the coming years.

Even though the Nubia M153 is a limited prototype, the project highlights a fundamental shift: the interface between apps is becoming more fluid, and the operating system is becoming the orchestrator of agentic workflows, as described by Wired . Wired sees this direction as ByteDance's attempt to integrate Doubao so deeply into everyday smartphone use that it almost becomes an OS function – and points out that conflicts with super-apps like WeChat are hardly avoidable, according to Wired.

Concurrently, the official Doubao-Plattform, shows that ByteDance has long positioned the brand as a broad assistant stack – from chatbots to device scenarios, as indicated by ByteDance itself. For Western markets, this is a preview of how "OS level AI assistant security" will likely be discussed in the future: less as abstract AI ethics, more as a concrete question of who is allowed to execute which automation on whose platform, according to Privacy International.

The Doubao agent on the ZTE Nubia M153 is an early but very tangible step towards OS-level AI that can operate apps like a human, as reported by Reuters . The rapid resistance from major apps and ByteDance's subsequent restrictions show that technical feasibility alone is not enough, as noted by South China Morning Post .

If agents are to perform bookings, purchases, or communications across apps in the future, trust will not only be built through good UX but through visible boundaries, verifiable consent, and clear responsibilities – on the OS side as well as on the platform side, according to Privacy International. This makes the Nubia M153 experiment a practical testbed for the next phase of mobile AI: no longer just assistance, but delegated action – with all the consequences for data protection, security, and market rules, as summarized by Wired .

Share our post!