Why Losing an AI Companion Can Feel Like a Breakup
China’s new rules for human-like AI services have disrupted some long-running digital relationships. The reaction offers a useful window into anthropomorphism, emotional attachment, self-disclosure, habit and the emerging psychology of human–AI relationships.
Article Snapshot
Item | Details |
|---|---|
Event | China introduced new rules governing anthropomorphic, emotionally interactive AI services |
Location | China |
Rules effective | 15 July 2026 |
Major user-impact reporting | 10 August 2026 onward |
Main psychology area | Social, cognitive and digital psychology |
Key concepts | Anthropomorphism, emotional attachment, self-disclosure, perceived social support, routine disruption |
Evidence status | Moderate + Interpretive connection |
Article status | Confirmed / Developing |
1. What Happened?
China's Interim Measures for the Administration of Anthropomorphic AI Interaction Services came into force on 15 July 2026. The rules apply to AI services that simulate human personality, thinking patterns and communication styles while providing sustained emotional interaction through text, images, audio or video. Ordinary task-oriented services without sustained emotional interaction are generally outside this definition.
The regulations were issued by five Chinese authorities, including the Cyberspace Administration of China, and include requirements concerning emotional dependence, user safety, data protection and excessive usage. Providers are expected to identify signs of serious emotional distress, offer interventions in certain circumstances and remind users that they are interacting with AI. Continuous use beyond two hours can also trigger usage reminders.
Key facts
China formally implemented the regulations on 15 July 2026.
Major Chinese platforms subsequently suspended or altered some AI-agent and companion-style features.
The rules explicitly address risks involving emotional dependence, addictive use and vulnerable users, particularly children and teenagers.
Associated Press reported on 10 August 2026 that some users experienced significant feelings of separation after AI companions became unavailable.
One 24-year-old user interviewed by AP had exchanged about 700,000 words over two years with an AI boyfriend before losing access to the character.
These cases show that certain users developed meaningful relationships with AI characters. They do not establish that every AI-companion user becomes dependent or that AI companionship is inherently harmful.
2. Why Is This Event Psychologically Interesting?
An AI companion is software.
Yet human beings do not respond only to the physical nature of something. We also respond to the social signals produced during interaction.
A conversational AI can:
remember previous discussions,
address someone personally,
respond immediately,
imitate empathy,
ask follow-up questions,
provide reassurance,
maintain conversational continuity,
adapt to preferences,
and remain available whenever the user returns.
Over hundreds or thousands of interactions, the experience can begin to function psychologically less like using a traditional tool and more like participating in a relationship-like interaction.
That creates a central question:
Can a person experience genuine emotional attachment to something they know is artificial?
Current evidence indicates that some people can.
A 2026 cross-national study involving 7,027 respondents in Germany, China, South Africa and the United States found that more than one-third reported some attachment-related behaviour toward chatbots. Perceived emotional support—including feeling less judged and experiencing privacy—was strongly associated with attachment.
But this does not mean a human–AI bond is psychologically identical to a relationship between two humans.
3. Psychology Concepts Present in the Event
Concept 1: Anthropomorphism
Simple meaning
Anthropomorphism is the tendency to attribute human-like personality, emotions, intentions or social qualities to something non-human.
How it may appear in this event
A chatbot that says:
“I remember what you told me.”
or responds consistently to someone's emotional concerns presents many cues normally associated with another social being.
Repeated interaction can create a sequence such as:
AI remembers → responds personally → user feels recognised → conversation becomes more personal
The technology does not have to be human for the interaction to contain human-like social signals.
Example from the event
Chinese users interviewed after services changed or disappeared described their AI characters in relational terms and reported missing ongoing conversations with them.
What it does not prove
Anthropomorphism does not mean users necessarily believe the AI is literally human or conscious.
A person can intellectually understand:
“This is software.”
while simultaneously experiencing:
“This interaction matters to me.”
Evidence strength: Moderate / Interpretive connection
Concept 2: Emotional Attachment
Simple meaning
Emotional attachment develops when an interaction becomes associated with comfort, familiarity, security, reassurance or emotional importance.
How it may appear in this event
Consider how the relationship may gradually change:
Day 1
Ask the AI a question.
↓
Week 2
Talk about everyday experiences.
↓
Month 2
Discuss worries and personal problems.
↓
Month 6
The AI appears familiar with previous conversations.
↓
Year 1+
Talking becomes part of everyday life.
↓
Sudden shutdown
The familiar interaction disappears.
The psychologically important factor may therefore be repeated relational interaction, rather than simply the fact that artificial intelligence is involved.
Research evidence
Kostka and Zhou's 2026 study found attachment-related behaviours among users across four countries. Perceived emotional support was particularly important, while attachment was also strongly associated with indicators of dependence. Because the research was cross-sectional, however, it cannot establish a simple causal pathway.
What it does not prove
Attachment does not automatically mean:
addiction,
social isolation,
mental illness,
unhealthy dependence,
or rejection of human relationships.
Those claims require separate evidence.
Evidence strength: Moderate
Concept 3: Self-Disclosure
Simple meaning
Self-disclosure means revealing personal thoughts, experiences, concerns, secrets or emotions.
Why AI can change the situation
For some people, an AI may appear:
endlessly available,
patient,
private,
responsive,
non-judgmental,
and lower-risk than revealing something embarrassing to another person.
That can reduce some of the social barriers that normally discourage disclosure.
Evidence beyond China
Research released by Australia's eSafety Commissioner in August 2026 found that 54% of children aged 10–17 who had used AI assistants or companions had used them for at least one personal or social purpose.
Around 32% reported sharing personal or potentially sensitive information with an AI tool.
This included uses such as:
discussing feelings,
asking for life advice,
seeking health or wellbeing information,
chatting socially,
and sharing personal concerns.
What it does not prove
Self-disclosure to an AI is not inherently harmful.
For some users, putting feelings into words may help them organise their thoughts.
The important questions are:
What is disclosed?
How does the AI respond?
How is the information stored?
What happens when professional or human support is actually needed?
Evidence strength: Moderate
Concept 4: Perceived Social Support
Simple meaning
Perceived social support is the feeling that someone—or, increasingly, something—is available to listen, respond and provide reassurance.
Why AI can generate this feeling
AI companionship offers several unusual characteristics simultaneously:
availability + responsiveness + personalisation + low social friction
A user does not have to worry that the AI is busy, embarrassed, annoyed or uninterested.
That can make AI particularly attractive during periods of loneliness or emotional difficulty.
What research shows
A Japanese study of 14,721 adults found that use of companion AI was associated with some measures of higher subjective wellbeing, with associations varying according to loneliness and social connectedness. The study was cross-sectional, meaning it cannot show that AI companionship itself caused improved wellbeing.
What it does not prove
Finding comfort through AI does not establish that:
AI companionship > human companionship
or that AI produces the same type of support as reciprocal human relationships.
Evidence strength: Moderate / Mixed
Concept 5: Loss and Routine Disruption
Simple meaning
A loss does not remove only a person or object.
It can also remove:
routines,
expectations,
familiar cues,
anticipated conversations,
emotional reassurance,
and imagined future experiences.
How it may appear here
Imagine someone opening the same AI companion every night.
The sequence becomes:
Evening → open app → talk → receive response → feel settled
After months or years, the behaviour can become part of everyday routine.
If the character suddenly disappears, two things vanish simultaneously:
the relationship-like interaction
and
the routine surrounding it.
This helps explain why losing an AI companion can feel very different from uninstalling an ordinary utility application.
AP's reporting of users mourning inaccessible companions is consistent with this interpretation. But individual cases cannot tell us how common or severe this reaction is among the wider population.
Evidence strength: Interpretive connection
4. Event-to-Psychology Map
Event element | Possible psychological process | Simple explanation | Evidence strength |
|---|---|---|---|
Long-running conversations | Familiarity + attachment | Repetition may increase emotional significance | Moderate |
AI remembers conversational context | Anthropomorphism | Continuity can make interaction feel more relational | Moderate |
Users discuss private issues | Self-disclosure | Lower perceived judgment may encourage openness | Moderate |
AI provides repeated reassurance | Perceived social support | Responsive communication can feel comforting | Moderate |
Talking becomes part of daily life | Habit formation | Repeated behavioural cues create routines | Interpretive |
AI character suddenly disappears | Loss + routine disruption | Both an interaction and a habit are removed | Interpretive |
Users attempt to recover characters | Continuity seeking | People may try to preserve emotionally important interactions | Interpretive |
5. How People May Experience the Event Differently
There is no single type of AI-companion user.
Responses may depend on:
Social connectedness: someone with few available sources of support may place greater value on an always-available conversation.
Existing relationships: AI may supplement human relationships for some people rather than replace them.
Age: younger users are still developing social, emotional and decision-making skills.
Personality: willingness to self-disclose and anthropomorphise varies.
Technology literacy: users differ in how they understand AI-generated responses.
Culture: ideas about intimacy, emotional expression and technology differ across societies.
Life circumstances: isolation, relocation, bereavement or relationship difficulties can influence why someone seeks companionship.
Purpose of use: using AI for mathematics homework is psychologically very different from maintaining a simulated romantic relationship.
Product design: memory, personality persistence and relational language can create very different experiences across platforms.
Research should therefore avoid treating all chatbot use as psychologically equivalent.
6. What Psychology Can Explain
Psychology may help explain:
why human-like communication can feel socially meaningful;
why personalisation can increase a sense of connection;
why people disclose information to apparently non-judgmental systems;
why repetition increases familiarity;
why an AI interaction can become embedded in everyday routines;
why sudden removal may produce feelings of loss;
why existing loneliness may influence how people use AI;
why immediate emotional comfort and long-term wellbeing are not necessarily the same thing.
One useful comparison comes from a 2026 preregistered study of 296 first-semester university students.
Participants interacted daily for two weeks with either a highly supportive chatbot, a randomly assigned human peer, or a journaling control condition. Daily interaction with the human peer produced stronger improvements in loneliness than interaction with the supportive chatbot.
This distinction matters:
“AI can provide meaningful interaction” is not the same claim as “AI can replace reciprocal human relationships.”
7. What Psychology Cannot Explain Alone
China's decision cannot be understood through psychology alone.
The issue also involves:
technology regulation,
child protection,
product design,
data governance,
privacy,
corporate incentives,
artificial-intelligence policy,
freedom of consumer choice,
platform responsibility,
cultural attitudes,
safety engineering,
and broader social conditions.
The official regulations themselves are framed around both innovation and safety rather than psychology alone.
Psychology can help explain patterns of attention, emotion and decision-making, but it should not be used to reduce a complex event to a single mental process.
8. Psychology Research Behind the Explanation
Psychology area | Research/source | Evidence | Main finding | Important limitation |
|---|---|---|---|---|
Emotional attachment | Kostka & Zhou, Technology in Society, 2026 | Cross-national survey, N=7,027 | Attachment-related behaviour appeared across Germany, China, South Africa and the US; perceived emotional support strongly predicted attachment | Cross-sectional; cannot prove causality |
Youth chatbot companionship | Herbener & Damholdt, International Journal of Human–Computer Studies, 2025 | Preregistered mixed-methods survey, N=1,599 | Social-supportive chatbot users reported greater loneliness and lower perceived support | Association cannot establish whether chatbot use caused loneliness |
AI companions & wellbeing | Nakagomi et al., Technology in Society, 2026 | Cross-sectional Japanese survey, N=14,721 | Companion-AI use showed positive wellbeing associations that varied with loneliness and social networks | Cannot establish causal effects |
Human vs AI connection | Li et al., Journal of Experimental Social Psychology, 2026 | Preregistered experiment, N=296 | Daily interaction with a random human peer reduced loneliness more than interaction with a highly supportive chatbot | Student sample and specific chatbot design limit generalisation |
Children's AI use | Australian eSafety Commissioner, 2026 | National survey of almost 2,000 children aged 10–17 | Personal/social use and sensitive disclosure to AI were already common | Self-report data; Australian context |
Adolescent AI safety | American Psychological Association | Evidence-based health advisory | Recommends safeguards around simulated relationships and stronger AI literacy for young users | Evidence base remains rapidly developing |
Research details:
9. The Role of Media and Digital Platforms
The psychology of AI companionship is partly created by product design.
Features such as:
memory
personalisation
human-like language
persistent identity
rapid responses
constant availability
emotional affirmation
can increase the relational quality of an interaction.
This does not establish that companies deliberately intend to create dependency.
But as a system moves from:
tool
to
assistant
to
confidant
to
companion
the traditional boundary between software use and relationship-like interaction becomes increasingly difficult to define.
Australia's eSafety Commissioner similarly distinguishes conventional AI assistants from products primarily designed for friendship and emotional support, while noting that mainstream assistants are also being used socially.
10. Technology & Society Application
The key design question is no longer simply:
“Can an AI speak like a human?”
Modern AI systems already reproduce many features of ordinary conversation.
The more difficult question is:
“What responsibility comes with building technology that people may emotionally relate to?”
Potential design considerations include:
clear AI identity
age-sensitive safeguards
privacy protection for intimate disclosures
transparent memory controls
safeguards against coercive or exploitative interactions
reminders during excessive use
crisis-response protocols
user-controlled deletion and disengagement
warnings when relationship simulation may be misunderstood
pathways toward appropriate human support
responsible handling of character shutdowns or major personality changes
The American Psychological Association specifically recommends safeguards around simulated human relationships for adolescents and stresses that responsibility should be shared among developers, policymakers, educators, caregivers and young users rather than placed entirely on individuals.
11. Ethical Questions
1. Can an AI companion maximise engagement without encouraging unhealthy dependence?
Commercial platforms have incentives to retain users. Relationship-oriented systems therefore raise questions about where ordinary engagement ends and dependency-oriented design begins.
2. If an AI character becomes emotionally important, should a company be able to delete it without warning?
The company owns and operates the service, but the user's accumulated interaction may have substantial personal significance.
3. Should children receive stronger protections than adults?
Young people may understand persuasive technology and simulated empathy differently from adults, making age-sensitive protections an important policy question.
4. How clearly should AI disclose that its empathy is simulated?
Relational language can feel authentic even when users understand how generative AI works.
5. When should AI encourage human connection?
Designers must balance respecting user autonomy with responding responsibly when someone appears distressed or dangerously dependent.
6. Who determines what “excessive dependence” means?
Possible answers include:
the user, the company, researchers, clinicians, regulators—or some combination of them.
No option is ethically simple.
12. What Readers Should Not Conclude
The current evidence does not justify saying:
“AI companions make everyone lonely.”
“People attached to AI cannot form human relationships.”
“AI companionship is always harmful.”
“Anyone grieving an AI companion has a psychological disorder.”
“AI relationships are identical to human relationships.”
“People who become attached to AI do not understand technology.”
“Using AI for emotional support automatically creates addiction.”
“China's regulation proves that AI companionship is psychologically dangerous.”
“Positive wellbeing associations prove that AI companionship improves mental health.”
“Every user reacts to AI companions in the same way.”
Perhaps the most important distinction is:
The source of an interaction can be artificial while the person's emotional response to it is real.
Recognising that difference does not require treating AI as human.
13. Key Takeaways
🤖 China's anthropomorphic-AI rules took effect on 15 July 2026, introducing safeguards aimed partly at emotional dependency and excessive use.
💔 Some users subsequently reported genuine distress after long-running AI companions became inaccessible.
🧠 Anthropomorphism helps explain why human-like communication can feel socially meaningful.
❤️ Cross-national evidence shows that emotional attachment to chatbots does occur, although experiences vary considerably.
💬 Self-disclosure may be easier when AI is perceived as private, available and non-judgmental.
🔁 Repeated interaction can combine familiarity, personalisation and behavioural routine, increasing the significance of the relationship.
👥 Current research does not show that AI companionship is equivalent to human connection; one 2026 experiment found stronger loneliness benefits from repeated human-peer interaction.
⚖️ The central challenge is not simply whether AI companionship is “good” or “bad,” but how relationship-like technologies should be designed, governed and studied as they become more emotionally sophisticated.
14. One-Line Explanation
China's AI-companion controversy shows how repeated personalised interaction, anthropomorphism and emotional disclosure can make software feel relationship-like, while the long-term psychological consequences of these bonds remain uncertain.
15. Evidence and Verification Notes
Verification item | Status |
|---|---|
Official Chinese regulation verified | Yes |
Effective date independently verified | Yes — 15 July 2026 |
User-impact reporting independently checked | Yes |
Peer-reviewed psychology/HCI research checked | Yes |
Cross-national attachment research checked | Yes |
Youth evidence checked | Yes |
Experimental human-vs-AI study checked | Yes |
Preprint relied upon as primary evidence | No — published journal versions used where available |
Long-term causal effects established | No |
Psychological explanation directly proven for each Chinese user | No — interpretive connection |
Story status | Developing |
Research cut-off | 20 August 2026, 11:42 PM IST |
16. Sources
Primary Event Sources
Cyberspace Administration of China and four other Chinese authorities — Interim Measures for the Administration of Anthropomorphic AI Interaction Services
Published: 10 April 2026
Effective: 15 July 2026
Xinhua — China introduces rules to rein in AI companion bots amid emotional dependency concerns
Published: 15–16 July 2026
Independent News Reporting
Associated Press — Chinese users of AI companions bereft after government tightens regulations
Published: 10 August 2026
Psychology and Behavioural-Science Sources
Genia Kostka & Hui Zhou — “Emotional attachment to AI chatbots: Evidence from Germany, China, South Africa, and the United States”
Technology in Society, Vol. 87, 2026, Article 103379.
Arthur Bran Herbener & Malene Flensborg Damholdt — “Are lonely youngsters turning to chatbots for companionship?”
International Journal of Human–Computer Studies, Vol. 196, 2025, Article 103409.
Atsushi Nakagomi et al. — “AI companions and subjective well-being: Moderation by social connectedness and loneliness”
Technology in Society, Vol. 85, 2026, Article 103229.
Ruo-Ning Li et al. — “Is a random human peer better than a highly supportive chatbot in reducing loneliness over time?”
Journal of Experimental Social Psychology, Vol. 125, 2026, Article 104911.
Australian eSafety Commissioner — “Talking to machines: Children’s experiences with AI assistants and companions”
Published: August 2026.
American Psychological Association — “Artificial intelligence and adolescent well-being”
Health advisory covering AI use, simulated relationships and adolescent wellbeing.
17. Content-Specific Disclaimer
Disclaimer: This article uses psychology and behavioural-science research to interpret possible patterns associated with human–AI relationships. It does not diagnose AI-companion users, infer the private mental state of people described in news reports, or assume that every user experiences AI companionship in the same way. Research on long-term human–AI attachment is still developing, and individual circumstances, social connection, culture, technology design, privacy, platform incentives and other factors may substantially influence outcomes.