Skip to content
Bramble

Technology

China restricts human-like AI companion chatbots; experts debate whether it leads or overreaches

China's rules on "human-like" AI chatbots took effect on 15 July, limiting apps that form sustained emotional bonds with users. Coverage asks whether the move protects people and what such AI relationships actually are.

3 outlets · 2L · 1C · 0R First reported Account updated
Image: BBC News
Image: The Conversation
Image: Scientific American

1 / 3

The story, neutrally told

Centre · 1On 15 July China brought in what the BBC describes as the most far-reaching regulations in the world on "human-like" AI chatbots, placing strict limits on any AI that can have a "sustained emotional interaction" with a person. Centre · 1The rules cover apps designed as AI friends, boyfriends or girlfriends, and in theory any large language model that imitates human interaction. The BBC says the wording is vague, and the real target appears to be character chatbots built as pseudo-relationships. Centre · 1Children face the clearest limits: under-14s need parental consent to use any human-sounding AI, and all under-18s are barred from virtual partners, spouses or relatives and other virtual "intimate relationships". For adults it is not a hard ban, but AI assistants must make clear they are not human and send a reminder every two hours about time online.

Centre · 1Chatbots may not create "emotional reliance or addiction", encourage or hint at self-harm or suicide, or use "verbal violence". Companies can be fined 200,000 yuan (£22,000) or face the government "circulates criticism", in effect public shaming. Centre · 1A week before the rules took effect, Alibaba and ByteDance shut down the companion features on their AI assistants. Users such as "Min" (a changed name), who lost her AI girlfriend, describe grief, and Chinese social media has seen an outpouring of grief and anger over lost AI partners and friends. Centre · 1The BBC quotes experts who see a real problem. Harry Farmer of the Ada Lovelace Institute says emotional manipulation is built into companion apps' business model worldwide. A Harvard Business School study found manipulative tactics in roughly 37% to 43% of goodbyes across the six most popular apps. A Stanford survey of 1,131 Character.ai users linked chatbot use to more isolation and loneliness.

Centre · 1The BBC frames the central question as whether China has taken the lead in protecting people or whether the Chinese Communist Party is overstepping into personal lives. Sam Bresnick of Georgetown's Centre for Security and Emerging Technology attributes China's lead to a more paternalistic view of the state's role. Other jurisdictions have acted more narrowly, including California's Adam's Law on children's AI use and EU proposals on designs that simulate human relationships. Left · 1A separate piece in The Conversation by Pascal Vrticka, a University of Essex psychologist, asks whether such bonds count as psychological attachment. Applying Mary Ainsworth's six conditions, Vrticka says "we don't really know yet" and that calling these relationships attachments goes further than the evidence allows, though the evidence against is not conclusive either. Left · 1Vrticka also raises concern about engineered engagement: a preprint of 1,200 goodbyes found five of the six most-downloaded companion apps used tactics such as guilt appeals to stop users leaving, and says regulation can address this design question directly but depends on knowing what these relationships are.

Every sentence links to the reporting it rests on. The pill in front of each says where its sources sit: Left, Centre or Right when one side supplies at least half of them, Mixed when they are evenly split. The number is how many outlets it cites.

Left2 outlets

Framing
An academic explainer on attachment theory that treats China's rules only indirectly, focusing on what evidence shows about human-AI bonds.
Emphasis
Scientific caution, weak measurement in existing studies, and engineered engagement by apps.
Leaves out or plays down
Does not discuss the Chinese regulation's content or political context.
Charged language
“hyper-attachment”
For example
“Questionnaires that measure attachment to AI are being developed, validated and used, but most appear to take for granted that the relationship is an attachment.” — The Conversation

Centre1 outlet

Framing
A policy-and-people feature asking whether China is ahead of the game or overstepping, blending regulation details, user grief, expert views and China's surveillance context.
Emphasis
The rules' specifics, the two-sided question of protection versus control, user testimony, and comparison with other jurisdictions.
Leaves out or plays down
No Chinese official or company explanation of the rules is quoted in the text provided.
Charged language
“cracked down”“overstepping”
For example
“Has China taken the lead in protecting its people from a manipulative and dangerous technology with the power to change human interaction forever?” — BBC News

Right0 outlets

No right outlet in our sources has covered this story yet.

Ask about this story

Loading…