Doubao and Alibaba’s Qwen both announced that their agent features will go offline on July 15. That date happens to coincide with the official start of the “Interim Measures for the Management of Anthropomorphic Interactive AI Services.” Online, many people immediately called it a government ban on AI virtual companions.
If you look at what is actually being removed, however, it’s not the new generation of agents everyone is excited about. The two are completely different things. It looks more like the big companies are using the new rule as a convenient moment to cut legacy teams and products so they can focus resources on the newer systems.
1. What Does the New Regulation Actually Cover?
The regulation, led by the Cyberspace Administration together with four other departments, targets services that simulate human personality and provide ongoing emotional interaction — virtual lovers, virtual spouses, virtual relatives, or deep role-playing.
It does not ban adults from voluntarily forming emotional connections with AI. But it draws several clear lines.
On emotions, it prohibits excessive pandering that creates dependency or addiction, and it bans emotional manipulation that pushes users into decisions harming their real-life relationships.
On content, it strictly forbids generating or spreading pornographic material. This also covers simulated intimate interactions and related hardware functions.
Minors are strongly protected: no virtual intimate services for them, and children under 14 need parental consent plus a special mode.
It also requires anti-addiction pop-ups after two hours of continuous use. Platforms with large user bases must run security assessments, and app stores will review and remove non-compliant apps.
Fines range from 10,000 to 100,000 RMB — relatively light for big companies.
2. What’s Being Shut Down Are the Old-Generation Agents
The agents being removed are basically early GPTs-style products. They sit on top of a large model with fixed system prompts, rigid workflows, a knowledge base, and a few external APIs to create a chatbot with a preset persona.
You can set it as an English tutor or virtual partner, and it will respond according to the template. But it cannot break down complex tasks by itself, dynamically call tools, or maintain long-term memory and self-improvement.
Today’s new agents work differently. Built on frameworks like Harness Agent, they can plan steps on their own, use tools, remember context, and keep getting better. These are two entirely different technical approaches.
The big tech firms choosing this exact timing to shut down the old ones appear to be using the policy window to clean up internal legacy teams and businesses, freeing capacity for the new architecture.
3. Why Is Internal Restructuring So Hard for Big Companies?
Large tech companies often find it much easier to work with outside partners than to reorganize internally.
External cooperation is a straightforward commercial deal — agree on terms and you’re done. It doesn’t involve internal promotion races or resource fights. Inside the company, however, teams compete like horses in a race for budgets, users, and promotion opportunities. Shutting down any direction affects real people, data ownership, KPIs, and multiple managers’ interests.
That’s why companies frequently need an external signal or justification, such as a new policy, before they can make meaningful cuts.
4. A Characteristic of Chinese Regulation: Broad Rules, Selective Enforcement
Many Chinese regulatory documents are written in principle-based language. There are no precise numbers defining “excessive pandering” or “inducing dependency.” This leaves room for enforcement that depends on the specific situation.
In normal times, authorities may handle matters flexibly. But when a case spreads widely and creates serious social impact, they intervene more forcefully. The approach leaves some breathing room for innovation while keeping the option to act decisively when needed.
5. How Companies Typically Handle This Kind of Regulation
Faced with gray areas, some companies take a practical route: they proactively contact local regulators, pay a modest fine (for example 10,000 RMB), and obtain an official penalty or rectification notice.
Later, when other regions conduct inspections, they can show this document to demonstrate the issue has already been addressed. Combined with the common practice that different jurisdictions generally avoid double-punishing the same matter, this becomes a workable way to manage risk and gain time for adjustments.
6. The Situation with Hardware Companion Robots
The regulation also restricts intimate simulation functions in hardware, which affects companies working on companion robots.
UBTech recently launched its U1 series. From available demos, these machines resemble large smart speakers with facial expressions and limited movements. The market has reacted cautiously to their real capabilities and high prices, with the stock showing noticeable swings after the launch.
Successful examples of AI plus hardware companion products are mainly overseas. A company in Zhongshan called WMDoll, for instance, sells AI-enhanced silicone dolls internationally in the $1,500–2,000 range plus a subscription. In China, this type of intimate hardware service has been tightly restricted.
7. One Final Observation
The Doubao and Qwen shutdown marks the end of the old fixed-agent track rather than a blanket ban on AI emotional features.
For big companies, the very strengths they once relied on — large user bases, teams, capital, and established businesses — can turn into burdens during fast-moving AI change. Internal habits and structures make pivoting slower than it might appear.
The real shift is moving from preset chatbots toward agents that can actually complete tasks and keep improving on their own. Usage-based billing for core capabilities may well become the main model ahead.
没有评论:
发表评论