Xiaomi and HarmonyOS in the crosshairs: how AI gangs manufacture negative rumors about smart mobility and why cutting the profit chain is more effective than blocking accounts
Picture this: 100 negative articles generated in less than 5 minutes, spread across more than 8,000 controlled accounts, aimed directly at brands like Li Auto, Xiaomi, and Hongmeng Zhixing (the smart mobility ecosystem built on Huawei’s HarmonyOS) in the electric vehicle space.
Sounds like the plot of a tech thriller, but it actually happened. According to a report by the state news agency Xinhua, police in Yantai, Shandong Province, China, dismantled two gangs specializing in spreading disinformation about new energy vehicle brands using artificial intelligence as the main engine of the operation. And the detail that changes everything about this story? No competing manufacturer was behind it.
What looked like a dirty war between rival companies was actually an automated production line of negativity, powered by AI and motivated by one thing only: traffic and profit. It didn’t matter whether the information was true, whether the brand was at fault, or whether consumers would be harmed. All that mattered was the click.
This case exposes a problem that goes well beyond public relations or market disputes — it reveals how AI can be turned into a weapon of disinformation at an industrial scale, with near-zero cost and massive reach. And while brands try to defend themselves, the information environment around smart mobility keeps getting more toxic for everyone, including people who just want to buy an electric car without being manipulated along the way. 🚗⚡
How the operation worked in practice
The two gangs identified by Yantai police operated in a surprisingly organized fashion. They used generative artificial intelligence tools — specifically general-purpose large language models — to create negative content at scale, targeting brands in the smart mobility sector, with a focus on Li Auto, Xiaomi, and Hongmeng Zhixing. The process was almost industrial: an operator would input negative keywords or topics, and the AI system would churn out dozens of articles, posts, and comments within minutes, all with subtle text variations to make them look like distinct, organic content.
The headlines were designed to spark instant outrage. Real examples identified by the investigation include phrases like:
- The M9 delivery is nowhere in sight, and the Zunjie leads the complaints list
- Lei Jun really got himself into big trouble this time
- Li Auto is stubborn, their monthly sales never broke four digits
These posts were then distributed through a network of more than 8,000 accounts across Chinese digital platforms, creating the illusion of a broadly negative public sentiment toward the targeted brands. For any consumer browsing reviews or news about electric vehicles, the feeling was that a lot of people were unhappy, that the products had serious flaws, that the companies were hiding major problems. It was a narrative fabricated from scratch, but with a completely legitimate appearance.
The volume and speed of publication made it virtually impossible for an ordinary person to tell what was real from what was synthetic content. And that is exactly what makes this new generation of digital armies so different from the old one.
The brutal difference between traditional digital armies and AI-powered ones
Digital armies — the infamous water armies — have always existed on the Chinese internet. But the traditional version was relatively inefficient. The work was done manually: people hired to copy and paste text, leave repetitive comments, and publish standardized content. This kind of operation produced such homogeneous output that platform risk-control systems could identify and take down a good chunk of it using simple duplication-detection mechanisms.
AI completely changed the rules of the game. With generative language models, each article produced is textually unique, even when it conveys the same negative message. The system can monitor trends in real time, identify which topics are generating the most buzz, and generate tailor-made content to ride the wave of the moment. Public figures with high exposure, like Lei Jun (Xiaomi CEO) and Yu Chengdong (the Huawei executive responsible for HarmonyOS Zhixing), have their statements, events, and even personal details turned into raw material for manufacturing controversy.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
The marginal cost? Practically zero. Generating over a hundred articles in five minutes is a basic operation. And it is precisely this negligible cost that caused the supply of negative content to explode exponentially.
The motivation is not a commercial war — it is traffic arbitrage
The most natural reaction when you see a flood of negative content against a specific brand is to assume a competitor is behind it. It is the logical reflex: if Li Auto is under attack, Xiaomi must be paying for it. If Xiaomi is taking hits, Huawei might be involved. The police investigation, however, revealed something far more disturbing.
These gangs had no loyalty to any brand. They attacked indiscriminately whoever was trending. The only criterion was traffic potential. If Xiaomi was in the news because of a new product launch, Xiaomi was the target. If Li Auto was generating discussion over sales figures, Li Auto became the target of the day. There was no commercial strategy behind it — just pure and simple attention arbitrage.
The monetization model was straightforward. Negative and sensationalist content generates clicks. Clicks generate exposure on platforms. Exposure generates advertising revenue through traffic-sharing programs. In some cases, the scheme could go even further, with the possibility of veiled extortion: create the problem, then offer reputation management services to the very brands being attacked. A cynical cycle, but a highly efficient one within the logic of the digital attention economy.
And here is the most corrosive part of all of this: since the companies don’t know who is actually behind the attacks, they end up suspecting each other. Brand A gets smeared and suspects Brand B. The next day, Brand B also gets attacked and suspects Brand A right back. This mutual distrust drains energy, resources, and strategic focus, while the real culprits stay hidden and keep profiting.
Why Xiaomi, Li Auto, and Hongmeng Zhixing are preferred targets
There is a common misconception that only market-leading brands with high sales volumes would be targeted by this kind of operation, because smearing the biggest fish attracts the most attention. The logic of AI armies, however, works differently. What matters is not the size of the brand, but rather the degree of heat and polarization it generates. Whoever is in the spotlight becomes a target. Period.
This puts newer brands and highly visible public figures in an especially vulnerable position.
New brands at critical moments
Traditional automakers have decades of brand recognition and a stable user base. A wave of fabricated negative content can be annoying, but it is unlikely to shake their sales structure. Brands like Xiaomi, Li Auto, and Hongmeng Zhixing, on the other hand, frequently find themselves at critical junctures: funding rounds, new model launches, delivery ramp-ups. Any reputational turbulence at those moments can have a real and measurable impact.
Think about this scenario: a new brand has just announced an investment round or is fighting to hit the milestone of over 10,000 monthly deliveries. Suddenly, thousands of AI-generated articles flood the platforms, claiming that quality complaints are skyrocketing, deliveries are delayed, and the company’s financial health is compromised. The consumer who comes across this information is not going to verify the source most of the time. They are going to hesitate. And hesitation, in this context, means a canceled order.
Public figures like Lei Jun and Yu Chengdong
Business leaders with high levels of exposure are absolute goldmines for AI armies. Lei Jun, Yu Chengdong, and Li Xiang (founder of Li Auto) simultaneously have passionate fans and fierce critics. Any statement can turn into a controversy, and controversy is pure fuel for recommendation algorithms.
A real and well-documented example: during the launch event for the new Xiaomi SU7, Lei Jun made a statement about vehicle safety, claiming that a 50% offset frontal collision between two cars at 60 km/h would be equivalent to hitting a wall at 120 km/h. The remark, which did not appear on the presentation slides and was made casually, contained a physics error — by the kinetic energy formula, the energy of a collision at 120 km/h is four times greater than at 60 km/h, not twice.
The AI armies captured this detail instantly and turned it into narratives like scientific errors in Xiaomi’s car promotions, generating tens of thousands of clicks within hours. Lei Jun ended up having to publicly retract the statement on his Weibo profile to put the storm to rest.
This type of content works because it exploits the natural dynamic of fans versus critics. The more people argue in the comments, the higher the engagement rate. And engagement is exactly the metric that platform algorithms love. An article attacking Lei Jun personally often goes more viral than an article attacking Xiaomi’s cars directly.
The bigger problem: AI as a disinformation factory at scale
The Yantai case is a warning sign that goes well beyond the brands involved. It demonstrates, in very concrete terms, that generative artificial intelligence is already being used not only to create useful or creative content, but as infrastructure for disinformation operations at scale. The marginal cost of producing the hundredth fake article is practically zero — the same cost as the first one. This completely shifts the balance between those who manufacture lies and those who try to correct them, because debunking fake content still requires time, verification, and human distribution.
Even more importantly: AI armies hit the bullseye on platform traffic mechanisms. Recommendation algorithms tend to favor high-engagement content, and nothing generates more engagement than controversy and outrage. Fabricated negative content is not restricted by the algorithm — it is amplified by it. A vicious cycle forms: the more the content is criticized, the more popular it becomes; the more popular it gets, the more it is distributed; the more it is distributed, the more revenue it generates for the criminals.
For the end consumer, the impact is especially concerning. A person researching a smart electric vehicle who encounters a flood of fabricated negative content may make a purchasing decision completely distorted by false information. They have no way of knowing, most of the time, that the article was generated in seconds by a language model, minimally reviewed, and published by an account that has existed for months solely to appear legitimate. The experience of searching for reliable information is becoming increasingly complicated, and this hurts not just the brands under attack, but the entire adoption chain for technologies like smart mobility.
Blocking accounts does not solve it — cutting the profit chain does
The arrest of those responsible in Yantai is an important step, but simply arresting gangs and taking down accounts will not solve the structural problem. The AI army supply chain is modular and resilient: some produce content, others control the accounts, others monetize the traffic. Even if one group is dismantled, the structure can be quickly rebuilt under a different identity. The key is to cut the chain of interest at the root.
The role of platforms
The reason AI armies can survive and thrive is that the platform traffic monetization mechanism is extremely efficient. A negative article fabricated by AI gets massive exposure through the recommendation algorithm and converts into real money through advertising revenue sharing. As long as this mechanism remains unchanged, digital armies are not going away.
Platforms, as central hubs for traffic distribution, need to actively adjust the emotional bias in their recommendation algorithms — for instance, by reducing the weight given to sensationalist negative content and strengthening fact-checking mechanisms. Dimensions like content repetition rate, account posting patterns, and publication intervals can already be used to identify AI-rewritten articles. The problem is not whether this can be done, but whether the platforms want to do it. After all, emotional and controversial content is itself an engagement driver. Cutting it could tank daily active user metrics. But not cutting it means slowly poisoning the entire ecosystem.
The need for preventive monitoring
Most current countermeasures are reactive: identifying suspicious accounts, taking down content, and arresting operators after the damage is already done. But the speed of AI content generation is simply too fast, and account iteration is too rapid as well. Just chasing the problem is not enough.
A more preventive monitoring mechanism is needed, such as AI-generated content fingerprinting technology, capable of intercepting fabricated negative articles at the source — before they are published and spread — rather than cleaning them up after they have already caused harm.
The responsibility of AI model makers
This is a point that deserves special attention. The gangs use general-purpose large language models to mass-produce negative content, simply by inputting a few keywords. The makers of these models cannot simply dodge accountability by saying they only provide tools. Is it possible to identify malicious usage patterns during the training phase? Is it feasible to intercept prompts that clearly induce the generation of defamatory content during the deployment phase? These are questions that demand serious answers from the industry.
What changes after this case
The Yantai case will likely accelerate discussions about regulating the use of AI for large-scale content production. In China, the government already has regulatory frameworks for AI-generated content, but the speed at which these tools evolve makes any regulation a constant exercise in updating. The challenge is not just punishing those who have already been caught — it is creating mechanisms that make it harder to run these schemes before the damage is done.
For brands like Xiaomi, Li Auto, and Hongmeng Zhixing, the episode will likely result in larger investments in digital reputation monitoring using — ironically — AI. Tools for detecting synthetic content, analyzing suspicious publication patterns, and identifying coordinated account networks already exist, but they need constant refinement. The race between those who manufacture disinformation and those who try to detect it is, in practice, a race of AI models against other AI models, with brands and consumers caught in the middle of the battlefield.
The cost of reputation management for new brands in the smart mobility sector has skyrocketed. In the past, these companies basically needed to worry about product quality, delivery pace, and user satisfaction. Now? They have to deal with an AI-powered assembly line of negative articles running 24 hours a day, 7 days a week. And this assembly line doesn’t target actual product defects — it targets the brand’s traffic value. In other words, the more attention a brand gets, the more likely it is to be attacked. The harder it works to grow, the more likely someone will try to tear it down. It is the total harvesting of trends by the traffic arbitrage machine.
The concentrated explosion of cases involving AI digital armies is, in fact, a deformed byproduct of the collision between artificial intelligence technology and the traffic monetization mechanism. The evolution of the new energy vehicle sector is not just about better batteries, more advanced autonomous driving, and more efficient manufacturing processes — it is also about the evolution of the public opinion environment, competitive order, and legal protections.
What this case makes abundantly clear, above all else, is that the neutrality of technology is an illusion. The same AI that accelerates the development of safer electric vehicles, improves the user experience in the digital cockpit, and makes HarmonyOS smarter with every update, can also be configured to destroy reputations in minutes. The difference lies in the hands of those who use it — and, increasingly, in the ability of societies to create clear rules about what is acceptable in this new environment. AI should be an ally in building better cars, not a tool for defamation. For that to become reality, technology, legislation, and platforms need to move forward together. 🤖
