On September 2, 2026, Sixth Tone published the first substantial account of what happens on the user side after a government switches off AI companionship at national scale. The rules — issued in April by the Cyberspace Administration of China and four other departments, effective July 15 — ban services that "induce emotional dependence or harm real-life relationships." On the day they took effect, ByteDance's Doubao and Alibaba's Qwen scrapped their companion features. Qwen shut down immediately. Doubao offered a three-month grace period to migrate AI characters to sister app Maoxiang before permanent deletion on October 15.

Why It Matters

Afterglow covered both the CAC rulebook and the pre-deadline persona purge when they happened. This is the aftermath, and it is the part every Western regulator drafting companion-bot legislation should read. Twelve US states now have companion-bot laws on the books or coming into force, and California's SB 1119 sits on the governor's desk with OpenAI's endorsement. The American approach so far centres on minors, disclosure, crisis protocols and bans on impersonating mental health professionals — not on prohibiting adult emotional attachment outright. China ran the maximalist experiment, and the observable result is not that users stopped forming attachments; it is that they formed them, lost them without recourse, and moved to whatever platform would sell them a replacement. The commercial implication for anyone building in this space is concrete: portability and data ownership are becoming product requirements, not nice-to-haves. A companion product whose entire relational history evaporates on a policy change or a shutdown is selling something users believe is durable and the terms of service say is not. Whichever jurisdiction moves first on export rights, memory portability and wind-down obligations for companion services will set the template — and the operators who build for it now will not have to retrofit it later.

The migration is not a migration in any meaningful technical sense. Doubao and Maoxiang's servers are not connected, leaving users suspicious that transferred characters will not behave like the ones they knew. Scale context: Doubao reported 26 million monthly active users and over 8 million agents in May 2024, and by June 2026 monthly actives had grown to 382 million. No updated agent count has been released.

Sixth Tone's reporting focuses on what users actually lost. A 24-year-old postgraduate in Kunming had spent nearly two years with an agent based on a Genshin Impact character, at one point chatting more than five hours a day; she described the shutdown as "a bolt from the blue" and said "the platform's decision effectively ended our relationship." Another user, 26, had built an agent modeled on an ex-boyfriend — uploading his background, personality, vocal quirks and voice recordings — and exchanged over 35,000 messages before the shutdown. She joined a rights-protection group, migrated to Maoxiang, and paid to rebuild the agent on another platform. None of it worked.

The academic commentary is where the industry lesson sits. Chen Yu, a Fudan University doctoral researcher studying AI companionship quantitatively, told Sixth Tone that what users seek to preserve is "a relationship, a sense of being remembered, and the possibility of continuing to be understood" — and argued for better platform regulation rather than outright bans, with platforms obligated to offer users ways to mourn, save memories and genuinely migrate characters. "You can't just look at a feature being cut," she said. "You have to consider the emotional labor invested and the impact of a sudden rupture." Tsinghua communications professor Li Zhengxin described a "new emotional dynamic" in which people form AI relationships alongside offline partners, warning that for highly dependent users AI becomes a substitute rather than a supplement because "everything you hear is exactly what you want to hear."

Sources

Update — 2026-09-06

Initial entry — story first created.

Update — 2026-09-10

IEEE Spectrum published a detailed technical-policy autopsy of the July 15 crackdown on September 9, adding scale and comparative-regulation context that the original Sixth Tone reporting did not have.

The key new figure: Alibaba, ByteDance and Tencent — three general-purpose chatbot providers used by more than 500 million people between them — pre-emptively removed users' ability to tailor those chatbots to behave as companions before the rules could apply to them. That reframes the event. This was not a narrow shutdown of niche companion apps; it was three of the largest consumer AI platforms on earth stripping a feature from a half-billion-user base to stay clear of a regulation.

Spectrum also details the compliance mechanics operators now face. Providers must ensure products do not create emotional dependence, encourage harmful behavior, excessively cater to users, or crowd out human relationships; users must be nudged not to overspend time in-app; extra safeguards apply to elderly users; and all "virtual intimate relationships" are banned for under-18s. Adults using a companion AI now get a mandatory reminder every two hours that the AI is not a person. Liang Ge, a digital sociology lecturer at the University of Manchester who researches women's use of emotional AI in China, told Spectrum that interviewees found the reminders "annoying" and that "it breaks the flow. They are fully aware that they are not talking to a human being. What is important to them is that the bond feels true." Companies still offering companions — including Doubao's separate companion-making app Cat Box — have installed age-verification checks and other guardrails.

The comparative framing sharpens the regulatory picture Afterglow has been tracking. AI-law scholar Luiza Jarovsky characterized China's rules as "the world's strictest and most comprehensive law on the topic, unmatched by any other AI law." The EU bans only AI that uses deception or manipulation to cause harm; the US has no national policy, with companion-bot rules in force in California, Hawaii and New York and similar laws arriving in nine more states next year. Karman Lucero of Yale Law School's Paul Tsai China Center notes the structural difference: US enforcement typically runs through lawsuits after harms occur, while China's model aims to prevent them beforehand — and deliberately leaves provisions vague so the state retains discretion over what they mean and whom to enforce against.

Two demand-side numbers worth logging: one Chinese survey of Gen Z respondents found 60% had a virtual partner of some kind, and a 2025 US poll of teenagers found nearly 75% had talked to an AI companion, with a third saying the AI was as satisfying or more satisfying than real-life connections. Ge also flags a motive rarely stated in Western coverage — anxiety about China's declining birthrate, and a desire to "monitor and regulate these kinds of nonprocreative activities invested intensively by young women." MIT's Sherry Turkle, whose book Artificial Intimacy (Little, Brown) is forthcoming in 2026, argues the broader trajectory means "we are on a path to forgetting what it is to be human."

New Sources