AI's true adoption by the masses does not depend on how much content it can generate, but on whether society trusts the content it produces.

Article author, source: ME News



TL;DR

• The Central Cyberspace Affairs Commission has launched the second phase of its "Clear and Bright: Rectifying AI Application Chaos" special campaign, focusing on combating AI-generated false information, vulgar and violent content, impersonation of others, and violations of minors' rights. To date, more than 5.61 million illegal and harmful pieces of information have been removed, over 49,000 accounts have been penalized, and more than 2,400 non-compliant websites and applications have been addressed.
• AI platforms such as Doubao, Yuanbao, Qwen, and Wenxin Yiyan are enhancing their training data moderation, model security filtering, and generated content labeling mechanisms—their core goal is not to limit AI capabilities, but to reduce the risk of technological misuse.
• AI governance is shifting from “removing违规 content” to “controlling the content generation process,” with model training, data sources, generation stages, and distribution channels becoming key regulatory focuses.
• The business competition in generative AI will, in the future, not only involve model parameters and product experience, but also security capabilities, compliance capabilities, and long-term trust.
• AI's true adoption by the masses does not depend on how much content it can generate, but on whether society trusts the content it produces.

The pace of artificial intelligence development is continuously reshaping public perceptions of the boundaries of technology.

Several years ago, discussions about AI focused more on model parameters, computational scale, answer quality, and commercial applications. Whoever had a larger model or stronger reasoning capabilities seemed poised to win the future.

But as generative AI enters the stage of large-scale application, a more pressing issue has gradually emerged: if AI can generate text, images, audio, or even entire sets of seemingly authentic information within seconds, who will ensure that these contents do not become new tools for spreading rumors, manipulating public opinion, or infringing on rights?

This is not a distant issue.

The Cyberspace Administration of China has recently launched the second phase of its "Clear and Bright: Rectifying AI Application Chaos" special campaign, directly addressing these real-world risks. This initiative focuses on cracking down on key issues such as using AI to create and disseminate false information, spreading violent or vulgar content, impersonating others, and infringing upon the rights and interests of minors. To date, the campaign has removed over 5.61 million illegal and harmful posts, suspended more than 49,000 accounts, and dealt with over 2,400 non-compliant websites and applications.

The issues behind these numbers are not due to AI technology itself, but rather the governance challenges that a rapidly growing industry must confront as it matures.

Any technology that changes the way society operates will not only improve efficiency.

The internet brought free flow of information, but also the spread of misinformation; mobile payments transformed financial experiences, yet gave rise to new fraud patterns; short videos lowered the barrier to content creation, but also fostered a competitive environment driven by traffic.

The same applies to AI.

It can enhance productivity but may also amplify low-cost illegal activities.

In the past, creating a fake article or a forged video required significant manual effort. Today, AI has lowered the barrier to content creation, giving ordinary users generation capabilities close to those of professional teams. Without appropriate governance mechanisms, the technological benefits could be exploited by a few to cause chaos.

Therefore, AI governance is not about putting the brakes on innovation, but about adding safety systems to a rapidly evolving road.

Many people may wonder: Isn't AI supposed to be as open as possible? Why are more and more AI products beginning to restrict generated content?

The essence of this issue is understanding the boundaries of responsibility for AI products.

Traditional software often merely provides tools, but generative AI is not a simple tool—it participates in the content creation process and can directly influence user perception.

When users ask AI questions, the content generated by the model may influence investment decisions, health choices, educational learning, or even perceptions of public events. Therefore, AI platforms must not only focus on “whether they can generate” content, but also consider “what is generated,” “why it is generated,” and “what impact it will have after generation.”

Platforms such as Doubao, Yuanbao, Qwen, and Wenxin Yiyan are strengthening content governance by establishing a new AI product security framework. According to notifications from the Cyberspace Administration, these platforms are strictly reviewing original training data, enhancing filtering of prohibited content, and reinforcing requirements for identifying AI-generated synthetic content to reduce the risk of illegal and non-compliant information at the source. (State Administration for Market Regulation)

This means that AI governance is no longer just a “reactive measure” from the traditional internet era.

In the past, platforms primarily responded to违规 content by discovering, deleting, and banning it.

But the AI era is different.

If the issue occurs during the generation phase, simply deleting the distributed content is not enough, because AI can continuously generate new variants, and the same type of违规 content may reappear in a different form.

Therefore, governance must be moved forward.

From training data to model capability control, and then to user-generated content moderation, forming a complete pipeline.

In short, the best AI products in the future won't just answer questions more accurately—they'll also know which questions to avoid, which content not to generate, and when to warn users about potential risks.

This is why AI security capabilities are becoming part of product competitiveness.

However, AI governance is not as simple as setting up a "block generation" wall.

If restrictions are too strict, AI products may lose their value.

Insufficient limits may also lead to risks.

This is a long-term game of boundaries.

For example, if a user wants AI to help write a novel involving elements such as war, crime, and conflict, does this constitute a violation?

A researcher wants AI to distinguish between analyzing cybersecurity vulnerabilities and malicious attacks.

A history enthusiast wants AI to adapt classic works—how can they tell the difference between reasonable creative interpretation and vulgar parody?

These issues demonstrate that AI governance cannot rely on simple keyword filtering.

Effective governance requires understanding the context.

This is why platforms are continuously enhancing their multimodal recognition capabilities, dynamically expanding their facial recognition, voiceprint, and violation sample databases. Platforms such as Douyin, Kuaishou, Weibo, Tencent, Baidu, Bilibili, Xiaohongshu, Zhihu, Douban, and Taobao are using technological means to improve AI content recognition and moderation capabilities, and have collectively issued 46 related governance announcements. (State Administration for Market Regulation)

The future AI security competition will likely resemble the autonomous driving industry.

A truly advanced system doesn't completely prohibit cars from driving, but rather helps them drive safely in complex environments.

The same applies to AI.

Mature AI is not a system that can do nothing, but one that finds a balance between open capabilities and risk control.

Some of the cases revealed in this special campaign also demonstrate new forms of AI misuse.

One typical issue is the use of AI to produce low-quality adaptations of classic works.

In the past, although online content creation was often poorly produced, production costs still limited its scale.

After the emergence of AI, batch generation became possible.

Some accounts use AI to radically alter classic works such as "Romance of the Three Kingdoms" and "Journey to the West," changing character settings and creating sensational plots to attract traffic. While these contents may seem like mere entertainment, their widespread and prolonged dissemination can affect the public’s understanding of cultural content and may undermine a healthy ecosystem of high-quality content.

Another issue is AI-generated misinformation.

In the past, creating a fake image or video required a high level of technical expertise.

Now, generative AI has lowered the barrier to entry.

False disaster scenes, fabricated social hot topics, and AI digital human marketing scams may all exploit public information asymmetry to create impact. The cases disclosed by the Cyberspace Administration in this notice include using AI to generate synthetic false disaster footage, fabricating social hotspot information, and conducting false promotions through digital humans.

More值得关注的是 AI 换脸换声。

Voice and facial appearance are important criteria for human identity verification.

When AI can simulate a person’s voice, expressions, and even behavioral habits, the identity trust system faces new challenges.

This involves not only public figures but also ordinary users.

Today, a person’s photo and audio recordings can both become sources of data that may be misused.

Therefore, data protection in the AI era is not only about preventing information leaks, but also about preventing the digital replication of personal identities.

Over the past few years, China's AI industry has developed rapidly.

From large model development to intelligent assistants, and further to application scenarios such as office work, education, e-commerce, and search, AI is transitioning from a phase of technological demonstration to practical implementation.

But the closer you get to ordinary users, the more important security becomes.

A business can accept that a model may occasionally give incorrect answers, but it struggles to tolerate a model that consistently generates false information.

A user can accept that AI has limitations, but it's hard to accept that AI helps create scams.

Therefore, future competition in the AI industry will not be solely about technology.

Parameter scale is certainly important, algorithmic capability is certainly important, but trust is equally important.

Why does the financial industry place such high importance on compliance?

Financial products connect funds.

Why does the healthcare industry place such a high emphasis on security?

Because medical decisions affect lives.

And AI is gradually connecting knowledge, information, and decisions.

It also requires building a foundation of trust.

From this perspective, this recent regulatory action by the Cyberspace Administration is not aimed at curbing AI development, but rather at guiding the AI industry toward a more mature phase. Clear regulatory boundaries, enhanced platform capabilities, and improved user awareness will ultimately foster a healthier ecosystem.

History has repeatedly shown that an industry truly matures not when it can create miracles, but when it can solve problems.

AI has proven its creativity.

It can help people write, code, and learn, and also enable businesses to improve efficiency.

But in the next phase, AI must prove something else: whether it is trustworthy.

Platforms such as Doubao, Yuanbao, Qwen, and Wenxin Yiyan are strengthening restrictions on违规 content, essentially addressing a key question:

As artificial intelligence gains increasingly powerful creative abilities, does it also develop a sense of responsibility commensurate with those abilities?

The decision determines not only the direction of development for several AI products, but also whether the entire industry can enter a long-term stable growth cycle.

The future of artificial intelligence does not belong to systems that merely generate more content, but to those that can find a balance between creativity, security, and social value.

Truly great technology is never defined by what it can do, but by what it knows it cannot do—and why.

1. Central Cyberspace Affairs Commission: "The Central Cyberspace Affairs Commission Actively Advances the Second Phase of the 'Clear and Bright·Rectifying AI Application Chaos' Special Action," September 2, 2026.
2. Central Cyberspace Affairs Commission: "The Central Cyberspace Affairs Commission has launched the 'Clear and Bright: Rectifying AI Application Chaos' special campaign," April 30, 2026.
3. Interim Measures for the Administration of Generative Artificial Intelligence Services, National Internet Information Office and other departments, 2023.
4. Measures for the Identification of Artificial Intelligence-Generated Synthetic Content, the National Internet Information Office and other departments, 2025.
5. Russell, Stuart & Norvig, Peter.Artificial Intelligence: A Modern ApproachPearson.
6. Shneiderman, Ben.Human-Centered AI. Oxford University Press.