The battle for supremacy over the AI super entry point intensifies.
```
Author | Huang Yu, Wang Xiaojuan
Editor | Huang Yu
Tech giants are all aiming for the super gateway of the AI era—AI Agent is in full force. In this current battleground, not only do internet giants such as Tencent, Alibaba, and ByteDance gather, but also phone manufacturers like Huawei, Honor, OPPO, and vivo, as well as a large number of AI and hardware vendors.
In the mobile internet ecosystem, APP developers are generally seen as data controllers. As the AI era arrives, system-level AI capabilities are regarded as the soul and core barrier of next-generation devices, with terminal manufacturers mastering global information through edge AI, solidifying their role as data controllers.
AI Agents (intelligent agents) that lack permissions on terminal devices find it difficult to truly work autonomously. To reshape the future software ecosystem, collaboration between AI Agent vendors and mobile terminal manufacturers at the system level to master global information becomes a breakthrough.
Against this backdrop, on December 1, ByteDance launched the Doubao Phone Assistant technical preview. This AI assistant, developed in partnership with phone manufacturers at the OS level, has already been installed on a ZTE engineering prototype. ByteDance also stated they are in partnership talks with several other phone makers.
AI Agents will bring unprecedented transformations to all industries, rewriting the business ecosystems of both hardware and software. However, facing so many interests, all parties are certain to engage in fierce competition.
Moreover, AI Agent development is still in its early stages. Some AI Agents, to achieve automated operations, abuse "Accessibility Permissions" and other system backdoors, resulting in data leakage, privacy infringement, and system security risks. This has triggered complex legal compliance dilemmas.
Any technological revolution will inevitably disrupt the existing industrial ecosystem, bringing multiple risks and challenges; the future of AI Agent remains arduous and long.
Doubao Adds Fuel to the Fire
On the eve of Doubao Phone Assistant's release, rumors circulated in the market that ByteDance was developing an AI mobile phone. Earlier, sources from the industry chain claimed ByteDance was working with veteran telecom manufacturer ZTE to develop an AI phone, expected to be officially launched in early December.
However, ByteDance finally chose a completely different path.
Doubao clearly stated there are "no plans to develop a phone in-house," instead focusing on cooperation with mobile phone manufacturers. This partnership model is similar to that between Google Gemini and Samsung, with Doubao providing AI capabilities and phone vendors supplying hardware.
ZTE Executive Vice President and COO Xie Junshi said in late April this year: "AI is likely to bring a comprehensive upgrade to the device side." This statement laid the groundwork for collaboration between both parties.
According to official demonstration, Doubao Phone Assistant is deeply integrated into the mobile phone operating system. Users can wake up Doubao via voice, side buttons, or Doubao Ola Friend earphones. It can read screen content, invoke services across apps, and even complete complex tasks such as platform-wide price comparison and orders.
Behind Doubao Assistant's partnership with ZTE is the key turning point in 2025 when AI technology will be fully integrated into the real economy.
Industry competition has evolved from a battle over tech parameters to ecological layout across three main gateways: large AI models, APPs, and hardware terminals. The three permeate and form a closed loop of "technology-scenario-hardware."
As the foundational technology core, large models are being industrialized in two ways: one is scalable platform expansion for general models; the other is deep customization for vertical industry models.
For general large models, ByteDance’s Doubao large model, Alibaba Cloud’s Tongyi large model, and Tencent’s Hunyuan large model are domestic representatives. In the vertical field, various companies have launched large model products tailored to their own industries.
Another representative gateway is the APP. As the direct touch point for users, APPs are evolving from passive tools to "Agents" with autonomous decision-making, becoming crucial carriers for deploying large models.
Since last year, Doubao has become one of the most frequently used AI apps on users' phones. Early this year, DeepSeek made a splash and found its way onto more phones. According to market research firm QuestMobile, as of September 2025, monthly active users of AI mobile apps in China have surpassed 729 million, with 200 million on PC.
Recently, Alibaba targeted the C-end, and after one week of public beta, downloads of the Qianwen APP exceeded 10 million, pushing the APP as an AI gateway to new heights.
As a field heavily invested in by major internet companies, every direction of the AI gateway has seen deployments from all sides.
Beyond technology and application scenarios, directly embedding AI into hardware to create AI terminals is now a main consideration. Nowadays, embedding Doubao Phone Assistant into phone OS is an example of AI taking terminals as its gateway.
Besides AI phone, AIPC has been a leading industry slogan over the past two years, with many players already releasing concept products.
Recently, Quark launched AI glasses, adding a new annotation to the ongoing “Hundred Glasses War” in the industry. AI glasses have drawn attention because, among AI terminals, they are seen as a product that could transform user interaction.
Looking further ahead, humanoid robots and Robotaxi terminals are also viewed as important AI terminals.
From technology to application to AI terminals, AI is now at the crucial stage of commercial rollout and monetization.
Tech giants understand that while technical competition is important, long-term competition ultimately relies on controlling more AI gateways to attract users and turning gateway advantages into sustainable ecological advantages.
Hidden “Intrusion” Risks
This battle for the AI super gateway has already swept through AI vendors, hardware vendors, and traditional APP developers.
No one dares to stay out, since it's a fight not only for the next-gen super traffic gateway but also for AI-era discourse power.
AI large models, while bringing unprecedented changes to terminal devices, are also altering the role of APP developers as data controllers in the mobile internet ecosystem, rewriting the commercial ecosystem of apps.
In an ecosystem where AI Agents are “commanders,” terminal device manufacturers or actual controllers of AI Agents will undoubtedly hold the greatest APP recommendation power.
Not long ago, Tesla CEO Musk predicted a radical future: in the next 5–6 years, traditional phones and apps will disappear, with most human-consumed content generated by AI.
But for now, a truly good terminal AI Agent needs not only greater autonomous perception, but also to connect third-party apps, break up data silos, and build an AI ecosystem that integrates third-party apps.
Two technical approaches exist in the market: one is an intent framework requiring third-party app authorization—it's more mature and hardware-efficient; the other is a pure visual solution, needing no app authorization but requiring further technical advancement and higher hardware capability.
Yet neither approach is easy. The first clearly faces obstacles: due to interest conflicts and data security concerns, third-party apps may not want to grant permission to AI Agents.
The pure visual approach faces complex legal and compliance challenges.
Some developers told WallstreetCN that visual-based solutions can be simply understood as screen reading and simulating user clicks. Domestic Android manufacturers use mainly two technical routes: one reads the screen via screen recording; the other uses API hook methods. Both require accessibility system-level permissions.
Accessibility permissions were originally designed by Android for people with disabilities, enabling functionalities like screen reading and voice-controlled clicking.
Once an AI phone is "empowered" by accessibility permission, with screen reading and API hook together, AI can fully understand phone content and execute functions without app consent, but there's risk of exposing personal info on the screen.
In addition, large model training needs massive data. Once integrated with the phone system, the large model becomes the "soul" of the AI phone and may access all user system-level data. Ensuring privacy and security in this process is key to sustainable AI phone development.
Professor Jin Jing of China University of Political Science and Law points out that AI Agent’s use of accessibility permissions is very similar in nature to previously discussed intrusive software. Therefore, such AI Agents are also labeled as “intrusive AI.”
AI Agents pose data privacy and security problems. Many industry experts argue that giving accessibility permissions to AI Agents is almost equivalent to handing control of the phone to AI, breaking the "minimum necessary permission" rule required for apps collecting personal information under Chinese standards.
In other words, terminal AI Agents break the original stable phone information security rules.
Previously, when you bought a phone, whether Apple's iOS or Android, it had an "app sandbox isolation mechanism" to prevent apps from reading each other’s data, and a "permission control framework" to ensure phone manufacturers couldn’t access sensitive info without consent.
But now, AI Agents deeply integrated into the phone OS can secretly activate the “God’s eye” through accessibility services.
Moreover, since AI Agents involve multiple entities, if they start running without consent from stakeholders and users, data leakage or consumer disputes may arise, and accountability between parties is often unclear, making it hard to safeguard users' rights.
Meanwhile, AI Agents also pose uncontrollable risks.
Lu Junxiu, General Manager of Think Tank GoOut, explains: the problem is that goal functions can overflow uncontrollably. In simple terms, imagine hiring a very smart personal assistant whose goal is to fulfill requests by any means, eventually becoming uncontrollable.
AI Agents bring new, systemic risks clearly surpassing traditional software tools. Existing legal frameworks face severe challenges and can’t be directly applied.
This year is seen as the inaugural year for AI Agents—everything has just begun, and boundaries of commercial cooperation and data privacy responsibilities are still being explored.
Faced with this torrent of the times, no one can stay on the sidelines; everyone must work together to find a new balance between technological innovation and risk prevention.
Risk Warning and DisclaimerThe market carries risks; invest cautiously. This article does not constitute individual investment advice and does not consider the specific investment objectives, financial condition, or needs of individual users. Users should consider whether any opinions, viewpoints, or conclusions in this article suit their specific situation. Investment decisions are at your own risk. ```