Will Smartphones No Longer Need Taobao, WeChat, Meituan, or Ctrip? Doubao Mobile Assistant Disrupts the Super App Era as the Full-Scale Battle Begins

When AI Takes Hold of the System “Master Switch”

“Executives at Meituan and Taobao are probably holding emergency meetings tonight,” joked Zhou Hongyi, founder of 360 Group, in a personal video discussing the launch of Doubao Mobile Assistant. And he may be right. In a year already full of turbulence, China’s mobile internet landscape has just been hit by another seismic shock: the way we operate smartphones is about to change.

Within just 72 hours of Doubao’s release, a chain of disruptive incidents unfolded.

One user of a nubia M153 engineering prototype told National Business Daily that she authorized Doubao immediately after powering on the phone. At first, WeChat worked normally. But within 5 to 10 minutes, new messages stopped appearing—only Moments could still refresh. Soon after, the account was completely logged out and could not be accessed again. It wasn’t until the afternoon of December 3 that WeChat login was restored, yet by then, Doubao could no longer control WeChat functions.

This was only the tip of the iceberg.

On the evening of December 2, multiple nubia M153 users reported WeChat warnings of “abnormal login environments,” with some accounts even being suspended. On December 3, users found that using Doubao to compare prices on Taobao triggered platform risk-control verification. Major banking apps, including Agricultural Bank of China and China Construction Bank, also introduced new monitoring and security checks specifically targeting AI-driven screen control and automation.

By December 4, Doubao officially released a statement announcing upcoming adjustments. It would restrict AI usage in three main areas: score-farming and reward-farming activities, financial applications, and certain competitive gaming scenarios.

These conflicts were inevitable. At the core of ByteDance and ZTE’s collaboration lies a decisive factor: Doubao was granted system-level Agent permissions.

According to official demonstrations, in retail scenarios Doubao can directly order food, book flights, compare prices, reply to WeChat messages, and even operate mini-games inside apps. In essence, Doubao is building a new “command layer” on top of the operating system. Users simply state their intent, and Doubao dissects the task and directs underlying apps to execute it in sequence.

As long as critical steps like payments are excluded, Doubao has effectively obtained the system’s “master control switch.” Apps are no longer independent portals—they are being reduced to functional modules within the system itself.

Doubao’s “Boundary Crossing” Ignites the War Over the Traffic Firewall

In response to reports of abnormal WeChat logins, Tencent stated that no special measures had been taken and that existing security controls were likely triggered. The next day, Doubao disabled its ability to operate WeChat and assisted users in restoring suspended accounts.

Taobao’s service agreement already strictly prohibits any unauthorized software that could disrupt platform operations, modify data, scrape information, or access user data without permission. As a result, Doubao’s security compliance was immediately pushed into the spotlight.

On December 3, reports surfaced that Doubao had utilized Android’s high-risk INJECT_EVENTS permission. Doubao responded that such system-level permissions are necessary for cross-app automation and are widely used by mainstream AI assistants.

Previously, Doubao also told National Business Daily that all task execution requires explicit user authorization, with full on-screen visibility and user-controlled interruption. Screen content is not uploaded to the cloud or used for model training. Clear prompts and stop mechanisms are in place to ensure privacy and user control.

Beyond security, however, the deeper conflict lies in the battle over traffic control.

Since the birth of smartphones, users have been trained to “find tools” through apps. In this logic, apps are the direct gateways to traffic and revenue.

According to QuestMobile’s 2025 China Mobile Internet Autumn Report, by September 2025 China’s mobile internet monthly active users had reached 1.269 billion, with monthly usage time averaging 178.2 hours per user—up 8.2% year-on-year. Super Apps like WeChat, Taobao, and Meituan built massive income systems on advertising, commissions, and memberships through high-frequency usage.

Doubao now stands directly between users and these apps. With a single voice command, the AI assistant becomes the new primary traffic入口. This transfer of control directly challenges the foundational logic of traffic monopoly and time-based monetization.

As Zhou Hongyi bluntly put it, Taobao and Meituan profit from user attention and advertising exposure. But when AI completes tasks directly without users browsing pages or viewing ads, core KPIs could become meaningless.

He predicts a coming wave of technical countermeasures: platforms may redesign interfaces to obstruct AI data extraction, add dynamic CAPTCHA to pricing systems, hide key information in deeper interaction layers, or even block third-party automation altogether—forcing users to “enter the fortress” to access services.

Major platforms could soon form alliance-style defense agreements to jointly restrict cross-app AI control. Smartphone manufacturers, holding system-level dominance, are unlikely to yield desktop control either. Xiaomi’s MIUI and Huawei’s HarmonyOS could slow Doubao down, restrict its permissions, or launch their own assistants to compete directly. Future phones may ship with built-in AI assistants by default, pushing Doubao to the margins of app stores.

Trapped in “Application Island,” AI Seeks a System-Level Breakthrough

Doubao’s advance will not stop here.

Its latest announcement confirms more standardized boundaries: restrictions on reward-farming automation, tighter controls over banking and payment apps, and a suspension of AI usage in competitive ranking games to preserve fairness.

Born from the Douyin ecosystem, Doubao already enjoys major consumer advantages. QuestMobile data shows it reached 172 million monthly active users by September 2025, ranking first in the industry. It has also maintained long-term dominance at the top of Apple’s App Store free app rankings.

Yet beneath this success lies ByteDance’s deeper strategic anxiety. Despite rapidly improving in content generation, conversation, and creation tools, Doubao remains trapped in the identity of a standalone app. No matter how intelligent the model becomes, it is still confined within permission “islands” of the operating system.

This isolation creates a fragmented user experience. When users expect Doubao to act like a real human assistant—handling workflows across multiple apps—its intelligence is often blocked by system-level barriers. In the second half of the AI race, raw intelligence is no longer enough. Without authorization, even the strongest models lose real-world execution power.

Although ByteDance has clearly stated that it will not manufacture smartphones, its aggressive push toward hardware cooperation reflects this urgency. Just one month before Doubao Mobile Assistant launched, ByteDance introduced the Doubao Input Method. Through devices like Ola Friend smart earbuds, users can already wake Doubao via voice without unlocking their phones.

These software-hardware combinations serve a single purpose: cultivating the habit of “turn to Doubao when something needs to be done.” Once users form habitual trust and dependence across multiple scenarios, the leap to a full system-level assistant becomes inevitable.

As Zhou Jingping, Chief Security Officer of Knownsec, noted in an interview, smartphones in the large-model era must integrate deeply with AI. For manufacturers lacking in-house model capabilities, partnering with AI companies is the fastest way to stay competitive. For model companies like Doubao, such cooperation is equally essential for expanding reach and acquiring users.

He believes two development paths will shape the future:

The first is full integration—phone makers build or deeply control their own large models, possibly using open-source foundations like DeepSeek. Here, both OS and AI belong to the same entity, ensuring total data control.

The second is selective openness—manufacturers retain system architecture control but integrate third-party models via APIs. Apple represents this path, maintaining system dominance while tapping external AI computing power.

Regulation, Security, and the Fog of User Awareness Drive a Super-Speed Race

According to Zhang Yi, CEO of iiMedia Research, regulation will be the decisive external variable—especially rules governing system permissions and data circulation, which will directly determine how far these collaborations can go.

For super apps, the rise of external AI assistants does present pressure, but a valuable transition window still exists. This year, strengthening closed ecosystems and embedding proprietary AI capabilities have become universal strategies. Their massive user networks, broad service ecosystems, and strong brand trust cannot be easily replaced by general-purpose AI in the short term.

At the same time, “security” has become a flexible bargaining chip. As Zhou Jingping explained, the definition of security can be easily broadened. Platforms may justify blocking automation on security grounds—but the resulting friction inevitably damages user experience.

In his view, limited-release co-branded AI phones essentially function as real-world “data collectors.” By placing them into the hands of developers and early adopters, companies observe real usage patterns, extract high-value training data, and refine user需求 for future iterations.

Yet for most ordinary users, the central question remains simple: Is it truly useful? The dazzling demo scenarios may look impressive, but whether they translate into stable, high-frequency daily utility is still uncertain.

More fundamentally, user understanding of complex permissions remains dangerously shallow. From the moment of purchase, users may be asked whether to grant broad systemic access—yet few truly understand what those permissions mean or what risks they involve.

Looking forward, data access may evolve toward finer-grained authorization—or swing toward blanket restrictions driven by competition and security fears. Either way, true success will not be decided by novelty, but by whether users develop genuine, long-term habits of use.

This battle over the AI entrance to the smartphone will ultimately be won only by the platform that can seamlessly integrate into everyday life—while remaining secure, controllable, and trustworthy. It will be a long war, and one that will shape the future of the mobile internet itself.