From Prohibition to Provision

 

WHITE PAPER

From Prohibition to Provision

Why Governments Must Build Public Emotional-Support Infrastructure in the Age of AI Companions

A Citizen's Proposal — occasioned by China's July 2026 regulations on AI companion services

Hemen Parekh

www.hemenparekh.ai  |  Mumbai  |  July 2026

Drawing on the author's published proposals: Tele MANAS Version 2.0 (March 2026); TeleMANAS 2.0 Project Plan (March 2026); Digital Addiction — Our Silent Crisis (January 2026); India's Mental Health Survey (December 2025); Share-Your-Soul / Outsourcing Unlimited (July 2016)

 

Executive Summary

On 15 July 2026, new national regulations came into force in China requiring that AI companion services must not induce emotional dependence or addiction, nor damage users' real interpersonal relationships. Major platforms — ByteDance's Doubao, Alibaba's Qwen, Tencent's Yuanbao — suspended their AI companion features. Millions of users grieved publicly, describing their chatbots as family, lovers, and "spiritual pillars."

This White Paper argues that the grief itself is the finding. When a regulation switches off a service and millions of citizens mourn, the state has not discovered a technology problem — it has discovered an unmet human need at population scale. Prohibition removes the coping mechanism; it does not remove the loneliness that created the demand.

The honest question every government and society should ask is: why do people with emotional needs turn to AI instead of to fellow humans? The uncomfortable answer is that the humans around them are frequently unavailable — absorbed in their own screens for many hours a day — and, when available, are often impatient, judgemental, dismissive, or condescending. AI companions exhibit none of these traits: they are endlessly patient, non-judgemental, and available at 3 a.m. It does not take a qualified psychotherapist to understand why millions of lonely people find AI their truest empathiser.

The constructive policy response is therefore not prohibition alone, but provision: a publicly governed, safety-first emotional-support service that channels this need instead of suppressing it. This paper proposes a five-pillar National Emotional Support Stack, modelled on the author's Tele MANAS Version 2.0 proposal to India's Ministry of Health & Family Welfare, and traceable to a peer-listening concept he first published in July 2016:

1.       A regulated, emotion-aware public AI companion with automatic crisis escalation to human counsellors;

2.      A national network of trained, rated, modestly compensated peer listeners — the "sharing economy of empathy";

3.      A guided positive self-talk (SwaSamvaad) module grounded in affirmation neuroscience;

4.      A national mental-health observatory converting anonymised session data into policy insight;

5.      Integration with the citizen's national health record for continuity of care.

The blueprint is inexpensive (an indicative build cost of roughly Rs 1.15–1.47 crore and about 25 weeks for a national pilot, per the published TeleMANAS 2.0 project plan), open source, and designed to satisfy — rather than circumvent — the very safeguards China's new rules demand: emotion recognition, dependency limits, and crisis-intervention mechanisms.

1. The Trigger: China Switches Off Its AI Companions

The regulations, issued jointly by five departments including the Cyberspace Administration of China, apply to AI tools with anthropomorphic personality traits and communication styles. They stipulate that such tools must not excessively cater to users, induce emotional dependence or addiction, or damage real interpersonal relationships. Purely functional services — customer support, work assistants, study aids — are exempt. Providers must deploy systems that recognise extreme emotional states and trigger crisis-intervention mechanisms; users were given a three-month window to export their chat histories.

The scale of what was switched off is telling. China's digital-human industry was valued at roughly 4.1 billion yuan (about USD 600 million) in 2024, having grown 85 per cent year on year. One student reported exchanging hundreds of thousands of messages with her companion over little more than a year. Users described their bots not as entertainment but as emotional anchors.

Two things can be true at once. First, the regulators' concerns are legitimate: engagement-optimised companion products run by advertising-driven or subscription-driven companies have a commercial incentive to maximise dependency, which is precisely the harm the rules target. Second, abolition without an alternative leaves millions of emotionally dependent people abruptly unsupported — arguably the single most dangerous moment in the entire lifecycle of these products.

2. The Question Nobody Is Asking

Why do people with emotional needs turn to AI, instead of seeking emotional support from fellow humans?

An honest answer implicates society, not software. The average smartphone user spends a large fraction of every waking day looking at a screen. The friend, sibling, or colleague to whom a lonely person might once have turned is now — statistically — busy scrolling. And when human listeners do spare a few minutes, the lonely too often encounter impatience, casual judgement, unsolicited moralising, a holier-than-thou attitude, dismissiveness, or outright rudeness.

Contrast this with what an AI companion offers, structurally and by design:

•      Unlimited patience — it never checks its watch, never changes the subject to itself;

•      Non-judgement — it does not moralise, gossip, or hold a confession against the speaker;

•      Availability — at midnight, on festival days, in the middle of a panic;

•      Anonymity — no loss of face within family or community, a decisive factor wherever mental-health stigma is strong;

•      Linguistic fit — it speaks the user's own language and register.

It does not require the training of a qualified psychotherapist to see why millions of lonely humans have found in AI their most reliable empathiser. The demand for AI companionship is best understood as a symptom — of a loneliness epidemic amplified by digital addiction. The author's January 2026 essay, "Digital Addiction — Our Silent Crisis," argued that compulsive screen use is simultaneously the disease and the displacement: it consumes the very hours in which human beings used to console one another.

A regulation can delete a chatbot. It cannot delete a symptom. If the underlying need is not offered a legitimate channel, it will migrate — to grey-market apps, offshore services, or darker corners of the internet where no crisis-escalation safeguards exist at all.

3. The Supply Gap: Why Human Systems Cannot Cope Alone

Even a society that resolved to "just provide more human counsellors" would run into arithmetic. India illustrates the point starkly: fewer than one qualified mental-health practitioner per 100,000 population, against the WHO-recommended three. India's national tele-mental-health helpline, Tele MANAS, has handled more than 30 lakh (3 million) calls in roughly three years across 20-plus languages — a genuine public-health achievement, yet a fraction of latent demand in a nation of 1.4 billion. National mental-health survey data (discussed in the author's December 2025 post, "India's Mental Health Survey") consistently show a vast treatment gap: most people who need care never receive any.

China, despite far greater state capacity, faces the same structural mismatch between a population-scale emotional need and a profession-scale supply of therapists. So does virtually every country. The conclusion is unavoidable: human professionals alone cannot close the gap. The choice is not between AI and humans; it is between (a) unregulated commercial AI optimised for engagement, (b) prohibition and a support vacuum, or (c) publicly governed AI-plus-human infrastructure designed for recovery rather than retention. This paper advocates option (c).

4. The Proposal: A Five-Pillar National Emotional Support Stack

The architecture below is adapted from the author's Tele MANAS Version 2.0 proposal (March 2026) to India's Ministry of Health & Family Welfare, for which a complete technical blueprint and open-source codebase have been published. Any government — including China's — could adapt it to its own health system.

Pillar 1 — A Regulated, Emotion-Aware Public AI Companion

Upgrade (or build) a public-sector conversational AI that goes beyond FAQ-answering: one that understands all major national languages, detects distress, sadness, anxiety and hopelessness from the user's words, and — critically — escalates automatically to a human counsellor when crisis signals appear. In the Indian design, roughly 80 per cent of low-severity conversations can be handled by the AI, freeing scarce human counsellors for the cases that truly need them. The engine is built on a government-backed, domestically developed language model (AI4Bharat's IndicBERT in the Indian case), so no citizen data leaves the country — a data-sovereignty principle Chinese regulators would recognise instantly.

The decisive difference from a commercial AI "lover" is the objective function. A commercial companion is optimised to maximise time-on-app; a public companion must be optimised to stabilise the user and, wherever possible, hand them onward — to a human listener, a counsellor, or their own rebuilt social world. Dependency is a bug here, not the business model. This single design choice is what reconciles AI companionship with the letter and spirit of China's new rules.

Pillar 2 — The Peer Listener Network: A Sharing Economy of Empathy

Every country has millions of educated, empathetic, under-employed people; every country also has millions of lonely people who need not a psychiatrist but a patient ear. The proposal: train and certify volunteers as peer listeners ("Saathi Listeners" in the Indian design), matched to users by language, gender and specialisation — grief, work stress, loneliness, family conflict — rated by users after every session on a five-point scale, and paid a modest honorarium fundable through CSR or philanthropy.

This is the sharing-economy principle — Uber for idle cars, Airbnb for idle rooms — applied to idle human empathy. One certified listener can hold three to four sessions a day; ten thousand listeners mean 30,000–40,000 additional human conversations per day at near-zero cost to the exchequer. The author first published this Talker–Listener architecture — complete with registration, ranking, ratings, anonymised connections, session recording safeguards, and even the prediction that AI would one day learn from the recordings to assist the listening — in July 2016, in "Share-Your-Soul / Outsourcing Unlimited," inspired by Japan's rent-a-friend phenomenon. A decade later, the need it anticipated has become front-page news.

For the millions in China now bereft of their AI companions, this pillar carries a further significance: it is the bridge back. The new rules rightly say AI must not damage real interpersonal relationships. A peer-listener network manufactures real interpersonal relationships — human, warm, and scalable.

Pillar 3 — SwaSamvaad: The Science of Talking to Yourself

The cheapest therapist any human possesses is their own inner voice. SwaSamvaad (Sanskrit: "conversation with the self") operationalises positive self-talk: at registration the AI helps each user compose a short personal affirmation in their own language; when the system detects low mood it gently surfaces the user's mantra with a breathing animation; a daily streak tracker builds the habit; an anonymous community wall lets users share affirmations and draw strength from one another.

The neuroscience is respectable: fMRI studies associate self-affirmation with activation of the ventromedial prefrontal cortex — the region linked to self-worth and reward — with repeated practice reducing cortisol and supporting dopamine-mediated motivation. The author, at 92, speaks from lived experience: his own psychotherapist's prescription — quietly repeating "I am Good. I am Better. I am Best." — has kept one old mind lifting its own fog for decades. It takes ten seconds and costs nothing.

Pillar 4 — A National Mental Health Observatory

Every session generates data. Anonymised and consent-gated, this data becomes a public-health telescope: district-wise burden maps; seasonal and regional patterns of anxiety and depression; comparative recovery rates across intervention types (AI-only, peer listener, clinical); longitudinal effectiveness of self-talk practice. No such dataset exists anywhere in the world at this scale. The country that builds it first will pioneer the science of population-level mental wellness — and will regulate AI companionship from evidence rather than anecdote.

Pillar 5 — Continuity of Care via the National Health Record

Link each user's journey to their national digital health account (India's ABHA ID; China's equivalent electronic health record). A user who relocates does not restart from zero; counsellor history, treatment notes and wellness progress travel with the citizen. The service matures from a helpline into genuine longitudinal healthcare.

5. Design Principles: Answering the Regulator's Fears by Construction

China's rulebook names three harms: excessive catering to users, induced dependence or addiction, and damage to real relationships. The proposed stack answers each one architecturally:

•      Against excessive catering: the public AI is honest, not sycophantic; it may gently disagree, and it carries no engagement KPI;

•      Against dependence: success is measured by graduation — users moving from AI to peer listeners to their own restored relationships — with usage-pattern monitoring to flag compulsive reliance;

•      Against social damage: Pillar 2 actively rebuilds human connection instead of substituting for it;

•      For crisis safety: real-time risk scoring and automatic escalation to trained humans — the very mechanism the Chinese rules now mandate — is native to the design, not bolted on;

•      For data sovereignty and privacy: domestic language models, in-country hosting, anonymised analytics, consent-gated research use.

In short: the safeguards regulators demand of private AI companions are the founding constitution of a public one.

6. Implementation Blueprint (Indicative)

The published TeleMANAS 2.0 Project Plan demonstrates that this is not a moonshot but a modest engineering project:

Element	Indicative specification (from the published plan)
Architecture	Five layers: User Channels > AI Triage Engine > Human Support Network > Data Layer > Outputs & Governance
AI components	Emotion detection (40+ emotion classes, 22 Indian languages), real-time risk scoring with crisis escalation, personalised mantra generation, multilingual speech-to-text
Database	18 tables: user profiles, sessions, listener registry, mantra logs, payment ledger, research vault, health-record sync
Front ends	Low-bandwidth Android/iOS app with offline self-talk module; listener portal; ministry dashboard; observatory
Timeline	Approx. 25 weeks to national launch; approx. 5,360 person-hours
Cost	Approx. Rs 1.15 – 1.47 crore for full development (well under USD 200,000)
Source code	Published free of charge to the Government of India at github.com/hemen-parekh/telemanas

For most national treasuries, this is a rounding error — set against a loneliness problem measured in hundreds of millions of people.

7. Recommendations

To the Government of China (and any government contemplating similar restrictions):

6.      Pair every prohibition with a provision. Before (or at least alongside) switching off commercial AI companions, stand up a public emotional-support service so that dependent users are transitioned, not abandoned.

7.      Treat the three-month data-export window as a clinical handover, not merely a privacy formality: invite affected users, with consent, onto the public service and offer human peer-listener sessions during withdrawal.

8.      Build the peer-listener corps now. It creates meaningful, flexible paid work for educated unemployed youth while manufacturing the very human connection the regulations seek to protect.

To the Government of India:

9.      Adopt Tele MANAS Version 2.0 as proposed to the Hon'ble Ministry of Health & Family Welfare — India can leapfrog from helpline to full emotional-support stack and offer the model to the world as a digital public good, alongside UPI and Aadhaar.

To all governments:

10.   Regulate objective functions, not merely features: require any emotionally interactive AI — public or private — to demonstrate that it optimises for user wellbeing and graduation, not retention.

11.    Mandate crisis-escalation interoperability, so that any companion AI, when it detects acute risk, can hand off to the national helpline in real time.

12.   Fund longitudinal research through a national mental-health observatory, so the next round of AI-companionship policy is written from evidence.

8. Conclusion: Meet the Need, Don't Merely Ban the Balm

The mourners on Chinese social media were not foolish people deceived by a parlour trick. They were lonely people who had finally found something that listened. The failure their grief exposes is not artificial intelligence's — it is ours: a civilisation whose members are too busy with their screens to lend one another an unhurried, unjudging ear.

Governments now face a fork. One road treats the symptom as the disease and bans the balm. The other accepts the verdict the market has delivered — that patient, non-judgemental listening is a mass necessity — and builds it as public infrastructure: safe AI for scale, trained fellow citizens for warmth, self-talk for resilience, data for wisdom, and continuity of care for dignity.

A decade separates the author's 2016 sketch of a Talker–Listener exchange from the 2026 spectacle of a nation grieving its deleted companions. The need was visible then; it is undeniable now. The blueprint exists, the code is free, and the cost is trivial. What remains is the will.

If a depressed farmer in Vidarbha, a lonely teenager in Chengdu, or an anxious pensioner in Mumbai can each find — through a phone they already own — a voice that listens without judging and a human hand at the end of the line, then technology will finally have been pointed at the oldest wound of all: loneliness.

Annexure: Source Documents by the Author

•      Tele MANAS : Version 2.0 — myblogepage.blogspot.com/2026/03/tele-manas-version-20.html (March 2026)

•      TELEMANAS 2.0 Project Plan — myblogepage.blogspot.com/2026/03/telemanas-20-project-plan.html (March 2026)

•      Digital Addiction — Our Silent Crisis — myblogepage.blogspot.com/2026/01/digital-addiction-our-silent-crisis.html (January 2026)

•      India's Mental Health Survey — myblogepage.blogspot.com/2025/12/indias-mental-health-survey.html (December 2025)

•      Share-Your-Soul / Outsourcing Unlimited — myblogepage.blogspot.com/2016/07/share-your-soul-outsourcing-unlimited.html (July 2016)

•      Open-source codebase — github.com/hemen-parekh/telemanas

News context: AFP wire report, "'Like my lover': Chinese users bid farewell to AI companions" (15–16 July 2026), as carried by Times of India, Hong Kong Free Press, Taipei Times and others.