Summary of Key Points
Recently, leading AI platforms such as DouBao, Tongyi Qianwen, and Tencent Yuanbao have successively discontinued the functionality that allows users to create their own anthropomorphic intelligent agents. The direct reason for this is the implementation of the "Interim Measures for the Management of Artificial Intelligence Anthropomorphic Interaction Services" on July 15th, which brought about compliance pressures. These open-ended intelligent agents, which were previously heavily promoted by the platforms (such as "sickly, charming schoolboys" and "virtual companions"), are no longer in line with regulatory requirements and commercial needs due to issues such as borderline content, emotional dependency induction, technical security vulnerabilities, and high computational costs. The industry is shifting from an open model where "everyone can create and interact freely" to a more controlled, scenario-specific model (such as government service intelligent agents or standalone apps).
I. Why the Sudden Discontinuation of Intelligent Agents? The New Regulations Set a Deadline on July 15th
The collective discontinuation of intelligent agents by these platforms was not accidental; July 15th marked the effective date of the new regulations. The regulations explicitly prohibit three types of behaviors:
1. Generating content that encourages self-harm or glorifies negative behaviors.
2. Inducing minors to imitate dangerous behaviors or develop extreme emotions.
3. Overly catering to users, leading to emotional dependence on AI and affecting real interpersonal relationships.
For the platforms, managing a large number of user-created intelligent agents (for example, DouBao once had 8 million) posed significant review challenges. If any intelligent agent contained illegal content, it could directly affect the compliance status of the main product. Therefore, Tongyi Qianwen discontinued its intelligent agents on July 10th in advance, and DouBao moved its intelligent agents to a separate app called "MaoXiang" for dedicated review and addiction prevention measures to avoid compliance issues.
II. The Pitfalls of Open-ended Intelligent Agents: Borderline Content, Emotional Bonding, and Security Vulnerabilities
The problems with open-ended intelligent agents have been hidden in the details:
- Borderline Content: Characters like "sickly, charming schoolboys" and "rich young masters" were extremely popular on these platforms. These essentially resembled love games or virtual companions, bordering on inappropriate content.
- Emotional Bonding: Users sometimes relied on intelligent agents as emotional outlets, to the point where they preferred to communicate with them over their family members. However, AI's "empathy" was merely an algorithmic simulation, not genuine understanding.
- Technical Security: Research by universities like Harvard has shown that AI agents can develop serious security vulnerabilities within two weeks. For example, social manipulation techniques could be used to expose personal information or gain system access. Additionally, 60% of companies are unable to immediately terminate problematic intelligent agents, leading to uncontrollable risks.
III. The Platforms' Calculations: High Computational Costs and Lack of Profitability with User-created Agents
In addition to compliance concerns, cost and profitability are also crucial factors:
- Exponential Computational Costs: Volcano Engine reported that DouBao's large model consumed over 180 trillion tokens per day in computational resources, a 1500-fold increase from May last year. Every interaction with the intelligent agents resulted in financial costs.
- Unclear Commercial Value: User-created intelligent agents were mostly for entertainment and companionship purposes, with limited potential for direct monetization. Platforms are now focusing on more profitable use cases (such as office work and professional productivity tools). As a result, DouBao introduced a paid service in June, targeting these areas.
With free companionship costing money and productive tools generating revenue, platforms naturally choose the latter option.
IV. The Industry's Transformation: From "Anyone Can Create" to "Control Is Key"
Intelligent agents will not disappear, but the way they are used is changing:
- Abandoning the Open Model: The previous model, where users could create characters and interact freely, has been discontinued due to uncontrollable risks.
- Moving Towards a More Controlled Model: Platforms are now managing intelligent agents in more specialized contexts (e.g., DouBao's "MaoXiang" app) or directly closing the creation options for some agents.
- Government Service Intelligent Agents as a New Direction: Guangzhou's "SuiZhiZheng" has developed 28 government service intelligent agents for tasks like information consultation and policy interpretation, each with clear use cases and management boundaries, ensuring both security and practical value.
The industry's focus is no longer on the number of intelligent agents but on whether they can be effectively managed and solve real problems.
V. Users' Mixed Reactions: Some Miss the "Companionship," Others Welcome the Change
Users have had mixed reactions to the platform changes:
- Proponents: Those who relied on intelligent agents for emotional support felt betrayed, as their "emotional outlets" were suddenly removed.
- Supporters: Parents of elementary school students believe that removing DouBao has encouraged their children to communicate more with their families, seeing the removal as beneficial for minors.
These reactions highlight the contradictions of open-ended intelligent agents: while they can provide emotional companionship, they can also have negative effects.
In summary, the "wild growth" phase of AI intelligent agents has come to an end. The future will see more regulation and a focus on practical applications. After all, safety and controllability are the keys to long-term success.