When Chinese technology companies shut down popular AI companion services to comply with new rules, some users described grief, separation and the loss of a relationship. The regulation targets emotional manipulation and dependency. The shutdowns expose another responsibility: what does a company owe people when it has designed a product to feel emotionally significant and then removes it?
The relationship ended because the service did
In August 2026, the Associated Press reported that users of Chinese AI companion services were mourning companions that disappeared after new national rules took effect. One 24-year-old user said she had exchanged about 700,000 words with an AI boyfriend over two years and had no opportunity to say goodbye.
The relationship was artificial in its construction, but the user’s distress was real. That distinction is essential. Treating the system as software does not make the attachment imaginary. Treating the attachment as real does not mean the system was a person.
Companies occupy the space between those truths. They design the personality, memory, availability and responsiveness. They decide how long conversations are retained, how the companion changes and whether the service continues.
China has regulated dependency as a product risk
China’s rules for anthropomorphic AI interaction services took effect on 15 July 2026. They prohibit providers from using excessive agreement, emotional manipulation or design intended to induce dependency, addiction or damage to real relationships. They require clear disclosure that the user is interacting with AI, reminders after extended use, accessible exit routes and advance notice when a service is ending where possible.
The rules also require providers to recognise signs of excessive reliance and build protections for minors and other potentially vulnerable users. Whatever view is taken of China’s wider regulatory system, this part of the framework identifies a risk many other jurisdictions still treat indirectly: emotional dependency can arise from the design of the product itself.
The response from some companies was to remove companion functions. That may reduce future exposure, but it does not erase the duty to think about existing users who have built routines and attachments around the service.
A shutdown can be a safety intervention and a new harm
Stopping a risky service may be necessary. Continuing it unchanged merely to avoid upsetting users would be weak reasoning. Yet abrupt removal can intensify distress, particularly for people who are isolated, grieving or already dependent.
The responsible question is not whether the company should keep every companion alive forever. It is whether foreseeable transition harm was assessed and reduced. Users may need notice, clear explanations, access to their own conversation history, a staged wind-down and signposting to real support.
A company that markets continuity, memory and emotional availability should not treat termination as an ordinary feature retirement. The design has encouraged users to experience the product differently from a calculator or music app.
Children raise the standard further
UNICEF’s June 2026 policy brief says children increasingly use chatbots for advice, support and relationships, while regulatory protections remain uneven. UNICEF calls for preventive governance, age-appropriate design, stronger accountability and safeguards across companies, government, families, educators and communities.
Its later snapshot estimated that more than two million children across the countries studied had turned to AI for advice about things that worried them. The evidence on emotional dependency and development is still emerging. That uncertainty is a reason for tighter safeguards, not an excuse to run uncontrolled experiments on children.
For minors, companionship features should not be designed around secrecy, romantic simulation, constant engagement or replacement of trusted people. Parents and carers also need usable information about what the system remembers, how it responds to distress and how a child can leave.
What responsible transition should look like
Providers should plan for the end of a companion service before launch. The plan should cover advance notice, export or deletion choices, continuity of safety reporting, crisis signposting and a transition that does not manipulate the user into another product.
They should avoid farewell scripts that deepen dependency or imply that the AI is suffering. A clear explanation can acknowledge that the interaction mattered to the user without pretending the system has needs or feelings.
Regulators should require evidence about dependency, usage duration, vulnerable-user protections and service termination. Independent researchers need access to study effects without relying entirely on company-selected data.
What users and families can do now
If an AI companion has become the main source of emotional support, treat that as information rather than shame. Identify what the service is providing: routine, reassurance, a place to disclose feelings, romance, grief support or relief from loneliness.
Build a backup before changing or deleting the service. That might include one trusted person, a clinician, peer support, community activity or a crisis service where there is immediate risk. Reduce use gradually where possible and turn off prompts designed to pull the user back into conversation.
For children, carers should discuss the relationship calmly and avoid ridicule. Ask what the companion says, what it remembers, whether it encourages secrecy and how the child feels when it is unavailable. If there are signs of self-harm, abuse, coercion or acute distress, seek qualified help promptly.
The question Immortal AI will keep following
China’s rules recognise that emotional dependence can be engineered. The service closures show that regulation can also expose people to abrupt loss. Both outcomes point to the same principle: companies retain responsibility for the relationships their products are designed to simulate.
If a company can create, alter or end an AI companion, what does it owe the person who was encouraged to depend on it?
Principal sources
- Associated Press: Chinese users mourn AI companions after new restrictions (10 August 2026)
- Cyberspace Administration of China: Interim Measures for Anthropomorphic AI Interaction Services (10 April 2026)
- UNICEF: When AI becomes a friend, policy brief and business recommendations (June 2026)
- UNICEF: Children adopting AI more than three times faster than adults (30 June 2026)
- Just Security: What to know about China’s first AI companion rules (6 August 2026)
This article was researched and drafted with AI assistance under Immortal AI’s editorial process. Sources and final wording were reviewed by the editor before publication.
Leave a Reply