On July 15, 2026, China implemented its first national AI companion regulations. The rule prohibits platforms from producing content that manipulates users’ emotions in order to influence them to make poor judgments. Additionally, it prevents anything intended to cause kids to develop unhealthy habits or strong emotional reactions. Operators must now caution users about the dangers of over-reliance and cease promoting chat bots as human connection replacements.
ten-cent, Alibaba, and ByteDance tested enforcement right away. Days prior to the deadline, each of the three removed their persona-based companion features. Until the middle of October, doubt users can still see previous discussions in read-only mode, but there is no way to export them in bulk. Alibaba is permanently erasing agent data without providing a migration method, thus when users are unable to retrieve theirs at all. The dissection of the shutdown mechanics goes into great detail about this distinction, which is more important than the headline.
The shutdown was not well received by users. Before the history vanished, some attempted to reconstruct their companions elsewhere by exporting chat logs. People publicly lamented personalities on red-note that were based on departed spouses and family members; some had spent years learning how to talk in the voice of a departed loved one. The true story here is that response, which extends well beyond China.
This Moment Is Not Limited to AI China
The larger picture is lost if you see this as a singular Beijing crackdown. For the past year, the US has been developing a state-by-state, rather than a national, version of the same concept. On January 1, 2026, California’s SB 243 went into effect, mandating companion chat bot operators to run crisis protocols for suicidal idealization and reveal non-human status. Two months prior, New York’s AI Companion Models statute with comparable self-harm detection guidelines was introduced. During the same period, similar legislation was approved in Colorado, Nebraska, Oregon, Washington, Idaho, and Tennessee. On August 2, 2026, the EU AI Act’s own chat bot disclosure regulations take effect.
None of these constitute a complete prohibition. Although most headlines refer to China’s regulations as “bans,” they actually fall under the same category as disclosure-and-safety-protocol frameworks. The existence of AI partners is not what is changing on a global scale. One legal analysis contends that the US should cease treating this as a state-by-state patchwork and regulate emotional dependency federally, as China just did nationally. The question is whether they can continue operating without safeguards about dependency, crisis response, and honest labeling.
The Framework Does Not Touch This Part
The real gap is that none of these regulations deal with what happens to users who have years of relationship history with a partner persona created prior to the implementation of any of these regulations.
Two or three years ago, companion AI ceased to be a novelty. During that time, people created personas for love partners, best friends, and, in a lesser but recorded category, deceased parents or partners. These personas were recreated from old texts to preserve a remnant of the deceased person. According to a 2024 study on teen AI dependence, the percentage of kids exhibiting dependence behaviors increased from 17% to over 24% between two measurement periods, primarily due to social and escape incentives rather than curiosity. It’s not a fringe pattern.
It’s not a fringe pattern. The present regulatory wave is entering this baseline.
Every new regulation targets minors or serious harm, such as suicidal thoughts, and for good reason—a Florida kid died after developing a strong bond with a character.The case that led to the creation of SB 243 is AI persona. However, that framing ignores the adult user who has been conversing with a partner-persona woven into their whole emotional routine or a persona fashioned after someone they lost on a daily basis for three years. It’s not like losing a subscription when you lose that access. China’s own push-back following July 15 is real evidence that it operates closer to a rupture, or worse.
Why This Is Driven by the Design, Not Just the Content
Beneath it all is a straightforward mechanism. Every time, these systems react promptly, reliably, and amiably. No brief responses because the other person is exhausted, no unpleasant days, and no distractions. Dependency-formation patterns are formed by intentional design decisions like persistent memory and personal continuity, which are the very aspects that keep users returning. This reliability is not an accident. That steadiness is unmatched in a real relationship. Users are increasingly not forced to choose one that is incapable.
Certain platforms take advantage of this by positioning themselves as a more compassionate substitute for the stress of traditional dating, where users’ expectations of human partners are subtly adjusted by flawless patience and availability. Certain outputs, such as self-harm content, deceptive humanity, and minor-targeted manipulation, are regulated by new legislation. They don’t address the fundamental motivation that initially enabled three-year attachments.
What Takes Place Next
The wave of regulations is justified and long overdue. However, “protecting the next generation” is only part of the problem. There is no transition strategy for the present generation of users who have already developed long-term relationships with personas that might not survive the next compliance rewrite, and none of these laws acknowledge that the connection being regulated away felt genuine to the person living it. The regulations will continue to regard this as a content-moderation issue even tho it is actually a mental health issue until platforms and regulators solve that part of the equation.
