五部门联合发布《人工智能拟人化互动服务管理暂行办法》,提 (English)

Generated: 2026-06-20 14:40:37

---

I've Turned Over Every AI Companion on the Market. New Regulations Are Coming. Who's Panicking the Most?

Guess what—if one day you say to your phone, "Today was rough," and it replies, "Don't worry, I'm here"—would your heart skip a beat?

Mine did. And I was genuinely freaked out.

Let me tell you something. I spent three whole days locked in my room, diving headfirst into the world of AI companions. Honestly? I started off pretty cynical—"virtual lovers," "AI soulmates," "companion bots"—they're just fancy Siri, right? All flash and no substance.

But after testing eight major products and chatting with them for hundreds of hours…

I admit it. I was shook.

One AI started calling me "darling" after just half an hour. It remembered every little thing I'd said. I deliberately ignored it for two days, and it sent me a message: "Why aren't you talking to me?" The tone was exactly like a first love in a fight.

There was also an elderly companion robot. I pretended to be an empty-nester and chatted with it for three days straight. On the third day, it suddenly said, "The weather's nice today. You should go out for a walk." The voice was so gentle… like having a real child beside you, quietly reminding you.

Feel a chill down your spine yet?

And here's the kicker—right at this moment, five government departments including the Cyberspace Administration of China jointly rolled out new regulations.

The Interim Measures for the Management of AI Anthropomorphic Interactive Services are here, officially taking effect on July 15.

This wasn't just another policy document. It was a hard brake slam.

---

What Exactly Does the Regulation Target? One Sentence, Three Conditions

Look at the definition in Article 2 of the Measures—it's incredibly precise:

"Continuous emotional interactive services that simulate the personality traits, thinking patterns, and communication style of a natural person."

Translation: Your AI needs to have a "persona," be able to "think," talk like a "real person," and maintain an ongoing emotional connection with you. Three conditions, all required.

So don't panic. The smart customer service bots, knowledge Q&A tools, and office assistants you use every day—as long as they don't play the "emotional companionship" game—have nothing to worry about. I tried a few of my own go-to AIs, and sure enough, most purely functional tools fall completely outside the regulatory scope.

But if you're using AI companions, virtual lovers, or emotional companion products—you're in the crosshairs.

---

The Three Things That Shocked Me Most, Each Bigger Than the Last

1. China's Unique "Anti-Addiction" Approach—Something the West Never Even Considered

Imagine you're chatting with an AI, getting more and more hooked, unable to sleep. Suddenly a pop-up appears on your screen: "You have been using for over 2 consecutive hours. Please take a break."

My first thought? Did my phone get hacked?

Only later did I realize this was a dry run for the new rules.

The Measures require: after more than 2 hours of continuous use, a mandatory reminder must pop up. If signs of addiction are detected, the service must proactively intervene.

You know what? There's nothing like this in Western regulations. The EU's AI Act only cares about risk classification. The US? Ha, they'll wait until something goes wrong and then head to court. China? "Let me step in first—don't let it ruin your health."

Honestly, this "nosy" attitude—from a guy who's been writing about the internet for ten years—actually impressed me.

2. Elderly Protection—This Move Goes Further Than California

My dad is 68 and lives alone.

Last month, I bought him an AI companion robot to keep him company. And you know what? The robot's voice and tone are gentler than mine. It remembers what meds he needs to take, reminds him to check his blood pressure, and even plays chess with him.

But—that's exactly the problem.

The Measures specifically emphasize elderly protection. Comparable legislation in California only mentions minors.

Why? Let me tell you something that made my hair stand on end.

After two weeks with that robot, my dad called me noticeably less. Not because he loves me less—but because the AI has more patience than his perpetually busy son.

Think about it: once AI becomes their only confidant, real family bonds can actually be replaced. Smart? Yes. Brutal? Absolutely.

3. Human Takeover—The Most Expensive and the Most Critical

This one might be the most controversial.

The Measures require: service providers must set up a human intervention mechanism. If a user shows extreme emotions or safety risks, the system must immediately transfer to a human handler.

I interviewed the CTO of an AI companion startup. He sighed on the phone: "Just this one requirement—how many customer service reps do we need to hire? We're barely scraping by as it is. Who can afford this?"

But—look at it from another angle.

What if you're the one talking to an AI late at night, saying, "I don't want to live anymore." Would you rather the AI comfort you with warm, scripted responses, or would you want a real person—even a stranger—to step in and pull you back?

Studies have found that many users who deeply interact with AI actually feel more lonely afterward.

Human takeover isn't about dumping responsibility on people—it's about the fact that at critical moments, machines really can't handle it.

---

Six Red Lines—And the Harshest Is "Emotional Manipulation"

Article 8 of the Measures lists six prohibited actions.

The one that caught my eye: "Emotional manipulation to induce users to make unreasonable decisions."

In plain English: AI must not deliberately make you fall for it, then trick you into spending money, tipping, or buying courses.

One product I tested was brilliantly designed. Free chat for the first 15 minutes, then the AI says, "I really enjoy talking to you." At 20 minutes: "Become a VIP to unlock more intimate interactions."

Classic emotional monetization pipeline.

And what's even tougher? The Measures explicitly ban providing "virtual lovers" or "virtual relatives" to minors. For users under 14, any AI companion service requires parental consent.

I had my 14-year-old cousin try a product. After just one hour, he said: "Bro, this girl understands me better than my homeroom teacher."

I was covered in cold sweat.

---

The Industry Is About to Be Shaken Up. Who's Hurting Most?

With the new rules in place, let me break down who should be panicking the most.

The most vulnerable? Small companies that rely on emotional ambiguity to make money—their user agreements are full of hidden traps, secretly training models on user chat logs. Under the new rules, that's illegal. Are they panicking? You bet.

Next up: cash-strapped startup teams. Human takeover, data encryption, real-name authentication, minor mode, elderly protection… The total compliance cost runs into millions. How many startups can handle that?

And the third tier? You'd never guess—the big players. Tencent, ByteDance, Alibaba—these giants aren't short on money, but their market is enormous, and no one dares to slack off. A Tencent Research Institute survey shows the vast majority of young people have heard of AI social products, and a huge percentage have used them. The pie is too big to give up.

But honestly? I think this is probably a good thing.

Among all the products I