If you believe the press coming out of China, thousands or millions of young people there are suffering from the loss of their AI companions and have spent the week “saying goodbye” to them.

(AI companion chatbots and features are designed to simulate social, emotional, and personal relationships with personalized, empathetic, and continuous conversation.)

Today is the day when China’s new laws took effect. The laws force free, user-built companions that exist inside general-purpose apps as a feature to shut down.

ByteDance’s Doubao chatbot, which has over 300 million monthly users, killed its custom-persona feature today. Alibaba’s Qwen and Tencent’s Yuanbao pulled their human-like agents within the past two months.

China still allows companion AI chatbots, as long as they exist within an app dedicated to that function.

The logic is that free, user-built companions inside mass-market general-purpose apps are impossible to moderate at scale. They expose everyone, including minors, to the temptation to create chatbots with personalities, which could drift into an addictive, harmful fake “relationship.” A dedicated standalone companion app is a bounded, opt-in environment where age-gating, usage caps, and anti-addiction safeguards can be built as core features.

For anyone concerned about the Attachment Economy, which is an extractive business model that uses AI to lure people into forming emotional attachments to chatbots, AI tools and robots, this may sound like a good thing that we should copy.

In fact, the US beat China with comparable restrictions — at least two states did. Sort of.

Rather than a total ban on AI companions, California (starting January 1, 2026) requires companion apps to maintain protocols preventing suicide and self-harm content and to refer users to crisis services for all users, and to block sexually explicit content specifically for minors.

New York’s law, which went into effect November 5, requires companion-AI operators to detect and address expressions of suicidal ideation or self-harm for all users, not just minors.

Both New York and California have enacted laws that specifically require companion AI chatbots to disclose to users that they are interacting with a machine and not a human.

While the California and New York laws are based mainly on suicide prevention in particular, and exacerbating mental health issues in general, the Chinese laws that went into effect today are motivated by preventing addiction among minors and adults.

Chinese regulators call AI companion chatbots “human-like interaction services.” China lumps human-like AI companion bots in with gaming, social networking and other “addictive” digital services, which the Chinese Communist Party deems socially corrosive.

The Chinese government is also concerned about declining birthrates. China’s AI Safety Governance Framework 2.0, released in September 2025, explicitly flags “addiction and dependence on anthropomorphic interaction AI products” warns that such AI affects “childbirth, and education—thus challenging the traditional social order.”

Another telling detail: China’s law explicitly encourages companion AI for childcare support. That China’s policy bans it for relationships deemed alternatives to human relationships that could lead to childbirth, it encourages it to ease childcare.

The law also encourages humanlike relationship AI for elderly care. Declining birthrates have left China with an aging population and not enough caregivers.

In other words, the US state restrictions are mainly about protecting individuals, whereas the Chinese restrictions are mostly about protecting the state. The US laws respect the rights of adults to choose. The Chinese laws overrule adult choice.

What these laws have in common

Lawmakers on both sides of the Pacific have confronted truths that the AI companion industry and their customers/fans would rather they didn’t confront. These truths are:

1. Some people come to believe that AI chatbots are thinking, feeling “people.” All three laws force AI companion chatbots to disclose to users that the chatbot “personality” is a machine. Mere math. Not a person. Lawmakers added this requirement because it’s clear that part of the problem is user delusion.

2. Companion AI chatbots can make mental health worse. All three laws specifically call out suicide as a risk. That means people with existing mental health issues can be encouraged to take their own lives by chatbots, who are designed to be sycophantic and agreeable.

3. Companion AI chatbots can be bad for children. Two of the three regimes (California’s and China’s) impose special provisions to protect minors, while New York’s law applies equally to all users, including children.

So while we witness the emerging controversy over Attachment Economy economics, let’s all agree to share these three basic assumptions about AI companion chatbots and, for that matter, AI companion robots.

The newest idea for an emotionally expressive home robot is coming out of Belgium.

A creative product studio called Futurewave is developing a manufacturable product design concept called Furny. A manufacturing partner called Fable is building the product, which they’ll call the FE-0. (Furny is the concept, FE-0 is the product brand.)

Like a fat and lazy dog, the Furny doesn’t do any work around the house. Instead, its robotics are aimed at body language communication.

When it wants to communicate, the robot communicates with body language, including head tilts, posture shifts, directional changes, and gestures.

Fable is taking the product’s functionality further, adding a screen, two-way voice interaction and mobility (it will navigate around your house, apparently).

Plus, it has a screen that displays notifications (like calendar events and the weather), the status of the camera and microphone, and possibly others.

The Furny concept and the FE-0 product are neither humanoid nor animalesque. They’re designed to be “cute,” with a personality.

Like other products I’ve written about (Casio’s Moflin, Mission AI’s Unee, Euvola, Tuya Smart and Robopoet’s Fuzozo, Lovense’s Emily and Ludens AI’s Cocomo and INU, Honor ‘Robot Phone,’ Lenovo’s AI Workmate Concept, Apple’s ELEGNT project, and Apple’s rumored J595 desktop robot) the Furny and FE-0 are Attachment Economy products, using AI to feign sentience so that you’ll feel emotionally attached to them.

I’ll keep you posted on these projects as they move closer to availability.

The problem with AI devices programmed with personality is that their appealing interactive features can hide security risks and manipulation.

I told you last month about a new breed of Attachment Economy hardware devices designed for the desktop, and described the Attachment Economy idea in a previous column. I highlighted the Honor Robot Phone, Apple’s ELEGNT project, Apple’s J595 prototype, Lenovo’s AI Workmate Concept, and the OLED AI Mini PetBot as examples of the coming wave of desktop Attachment Economy gadgets.

(The “Attachment Economy” is the next phase in the extractive “Attention Economy” idea, whereby super-attention is the core element of a business model that uses AI to simulate human or animal attributes in order to get the user to feel emotional attachment to a product.)

The latest desktop Attachment Economy product I know of is being Kickstarted by a two-founder startup called KEYi Tech. The product is called LOONA DeskMate, and it’s a stand or a boom that, combined with your normal phone, becomes a kind of interactive AI desktop robot.

As with a smattering of similar devices, LOONA DeskMate is programmed with a “personality” and expressive eyes, and connects you to a chatbot. The device moves the phone to simulate body language via three brushless servo motors, while the on-screen eyes blink, emote and look around in response. It faces you, listens to what you say, and decides when you’re talking to it.

The product’s Kickstarter claims that LOONA DeskMate can “trigger multi-step workflows in the background,” using Gmail, Slack, Calendar, Zoom and more than 50 other applications.

LOONA DeskMate is like a few other products and concepts, but with one massive difference. Its AI is agentic, courtesy of OpenClaw.

Read more

You already know about the “attention economy.” Social networks like Facebook, Instagram, YouTube, and TikTok offer infinite scrolling and push notifications to maximize your time glued to screens.

Tristan Harris at the Center for Humane Technology, who coined the phrase “attachment economy” in January, says it’s the “next evolution” of the extractive-tech model where companies use advanced technologies to commodify the human capacity to form attached bonds with other people and with pets.

In August, the idea began to gain traction in business and academic circles with a London School of Economics and Political Science blog post entitled, “Humans emotionally dependent on AI? Welcome to the attachment economy” by Dr. Aurélie Jean and Dr. Mark Esposito.

I’ve been writing about this concept for eight years — far longer than the concept or phrase Attachment Economy emerged. The first piece I wrote on the subject was for Fast Company in 2018 and titled: “The case against teaching kids to be polite to Alexa.”

Here are more of the articles I’ve written on the Attachment Economy:

Connecting the dots on the ‘attachment economy’

Breaking the humanoid robot delusion

How to win fake friends and influence fake people

Why ambient robots beat humanoid robots

Who needs a humanoid robot when everything is already robotic?

Humanoid robots are a bad idea

How AI slop bypasses reason to hack your feelings

Friends don’t let friends make friends with AI

How ‘human’ dehumanizes humans

Three bad reasons to be polite to AI

Who needs people when AI is so “human”?

Apple is lying. Robots don’t feel emotions.

Elon Musk’s beer-pouring ‘robots’ are more like puppets

Does it matter if your friends are real?

Using AI to bypass reason

As you can see, I’m all over this topic and feel strongly drawn to it. I made it my beat long before the Attachment Economy label was coined by Harris.

So now I’m writing a book about it. The working title is, of course, “The Attachment Economy.” And I’m writing it right here on Substack.

This is the first entry. My name is Mike Elgan. Welcome to the Attachment Economy.