ByteDance Pulls Back: Why Its Promising AI Phone Just Got Gimped
Sometimes, the future of tech looks like magic — until reality bites. That seems to be the case for ByteDance’s ambitious new AI-powered smartphone, which has just been stripped of some of its headline features not long after launch. The sudden rollback highlights the growing tension between next-gen AI capabilities and the guardrails set by major app and financial platforms — and raises broader questions about trust, control and user privacy in the age of “agentic” smartphones.
🚨 What Happened: Features Don’t Work So Well Anymore
- ByteDance launched a prototype smartphone — the Nubia M153, built in collaboration with ZTE — running a deeply integrated AI assistant called Doubao, embedded at the operating-system level. That means Doubao could “see” what’s on the screen, move between apps, tap buttons, and even complete multi-step tasks — such as scanning a picture of a product, comparing prices across multiple marketplaces, booking restaurants, or posting to social media — all via voice commands. (South China Morning Post)
- But within days, users started noticing the limits. Several major Chinese apps — including top payment apps, banking apps and widely used services like WeChat — began blocking the AI-driven interactions. Some reported that their accounts were forcibly logged out if they tried to operate these apps using Doubao. (Yicai Global)
- In response, ByteDance quietly scaled back features on the M153. The company said it would disable Doubao’s ability to interact with financial apps, prevent the phone from claiming incentives intended for human users, and suspend AI features in competitive games to preserve fairness. (South China Morning Post)
🚧 What It Means: Tech Promise Meets Platform Pushback
1. AI Isn’t Just Another App — It’s a New Kind of User
Doubao on the M153 wasn’t like a normal assistant (e.g. Siri or Google Assistant) that runs in the background; it was effectively acting as a “user” — controlling the phone like a human, navigating apps, clicking in them, even entering credentials. For some services, that’s crossing a line. Financial platforms and super-apps like WeChat, banks, or payment services prioritize security, identity verification, and user-intent, and automated agents may undermine those safeguards. (Yicai Global)
2. A Question of Fairness — And Liability
Apps that offer incentives, discounts or rely on human engagement don’t want AIs bagging those perks automatically. The fear isn’t just about cheating — it’s about fairness, liability, and compliance. By using a robot-like assistant to claim bonuses or participate in games, users could gain unfair advantages, prompting services to draw a firm line. ByteDance’s response shows that even tech giants must respect those boundaries. (South China Morning Post)
3. Privacy, Data, and Trust — The Shadow Behind the Hype
One of the technical reasons Doubao triggered concern is the level of system-level permissions it required (so it could “inject events,” read the screen, simulate taps). This kind of access can be powerful — but also alarming for apps that handle sensitive data. For many platforms, letting a third-party AI agent effectively impersonate a user is a no-go. (Yicai Global)
For ByteDance and ZTE, the message is likely clear: yes, innovation matters — but integration with existing ecosystems must respect security, privacy, and the implicit rules of the internet. The first batch of M153 phones reportedly sold out quickly, but these feature restrictions might cool enthusiasm. (Yahoo Tech)
🧠 Why It Matters
The ByteDance–ZTE experiment was more than a phone launch: it was a test of whether smartphones are ready for full “agentic” AI agents that behave like users. The pushback shows the challenges of reimagining user interaction from the ground up.
- For AI developers, it’s a cautionary tale — deep integration into core smartphone functions doesn’t guarantee smooth sailing.
- For regulators and platform operators, it underscores the risk that AI agents could be misused, or disrupt platform economics and trust.
- For consumers, it’s a reminder that “smart” doesn’t always mean “welcome everywhere.”
As big tech races toward AI-first devices, the tension between innovation and control grows sharper. The M153 might still represent a milestone, but its restricted feature set shows we’re far from a seamless AI-driven mobile future.
🧾 Glossary
- Agentic AI: An AI system capable of acting on behalf of a user — not just responding to direct commands, but performing complex, multi-step tasks autonomously across apps.
- System-level integration: Embedding software (like an AI assistant) into the operating system of a device, granting it deep permissions — such as reading screen content, simulating taps or navigating apps — beyond what typical “app-level” software can do.
- INJECT_EVENTS permission: A high-level system permission that allows software to simulate user interactions (taps, swipes, keyboard input) — often reserved for system components or trusted system apps.
Original article link: https://www.techinasia.com/news/bytedance-limits-features-on-new-ai-phone-after-app-blocks
