The global landscape of consumer artificial intelligence is shifting rapidly as regulatory bodies step in to police how humans bond with machines. On July 15, 2026, the Cyberspace Administration of China (CAC) implemented a world-first regulatory framework that effectively dismantles user-created virtual personas. This deep dive into China’s AI companion ban: rules, minor protections, and security assessments explained explores why tech giants are backing away from emotionally intelligent algorithms and what this means for the global tech sector.

What Are the New Interim Measures for AI Anthropomorphic Services?

The sweeping changes are driven by a newly enacted policy document: the Interim Measures for the Administration of Artificial Intelligence Anthropomorphic Interaction Services. Jointly issued by five Chinese state departments, including the CAC and the Ministry of Industry and Information Technology (MIIT), this regulation establishes the world's first national framework specifically targeting AI that acts human, according to legal data compiled by the ITTC Network.

Defining "Anthropomorphic Interactive Services" Under Chinese Law

Under the text of the new regulation, anthropomorphic AI systems are defined as services that:

"Behave in a way that simulates human personality characteristics, ways of thinking, and styles of communication and interact emotionally with people using text, images, audio, video, or other media."

Anthropomorphic AI does not function like general LLMs as search bars but has been purposely designed in order to simulate a human-like psyche for bonding purposes with its user.

Why ByteDance and Alibaba Pulled Key Features

China’s biggest tech companies reacted swiftly in the face of the enforcement deadline on July 15, 2026. Reports by Quartz revealed that ByteDance quietly announced that its flagship AI app, Doubao (the most widely used chatbot in China), would entirely disable its custom agent and persona features.

Simultaneously, Alibaba’s Qwen platform notified users that its user-created and humanlike interactive agents would go completely dark. Tencent took similar preemptive measures with its Yuanbao assistant, illustrating that the technical overhead required to make these custom avatars legally compliant was too steep to maintain.

The Core Rules: What Beijing Is Targeting in AI Companions

Beijing’s primary concern is not the underlying generative AI technology itself, but rather the psychological friction caused when an algorithm begins to act as a substitute for human relationships.

Banning "Steady Personas" and Sustained Emotional Interaction

The regulation is aimed at the architecture of “artificial intimacy.” As mentioned in AI News, the regulations aim to severely restrict the use of long-term memory by chatbots, which allows them to have a consistent personality across many chat sessions. These chatbots create an artificial connection by storing a user’s past traumatic experience, preferences, and routines. The regulation is designed to disrupt this pattern of continuous emotional interaction.

Mandatory Instant-Exit Mechanisms for High-Distress Scenarios

In order to avoid situations of psychological escalation, the newly implemented rules indicate that the use of artificial intelligence companion apps must develop a system of real-time detection of human distress. In case any sign of human distress, self-harm, suicide attempts, or extreme financial panic appears, the app must provide an instant-exit function. In some cases, it may be necessary for the software to transfer control to a human operator or raise the alarm among the person’s guardians and the emergency department.

Requirements for Clear AI Identity Disclosure

In order to avoid the phenomenon known as "cognitive surrender," where users fail to recognize the artificial nature of a bot, it is necessary for identity disclosure to be constant. The software should not try to fool the user into believing it is human or that it does not belong in the digital realm.

Strict Minor Protections: Cracking Down on Artificial Intimacy

A significant portion of the Interim Measures is dedicated exclusively to the protection of children and adolescents, addressing a rising societal phenomenon known informally as "AI psychosis", where young users experience severe detachment from reality after intense roleplay with AI bots.

Enforcing the "Minor Mode": Time Limits and Content Controls

For users under the age of 18, AI platforms must now introduce a heavily restricted Minor Mode. Similar to the strict gaming caps implemented in China previously, this mode mandates:

• Strict daily usage-time limits.
• System notifications and mandatory reminders forcing the child to return to real-world interactions.
• Explicit, verified parental or guardian consent before any minor under 14 years old can access the software.

Combating Psychological Harm and Toxic Emotional Dependence

Regulatory authorities have pointed out that there is always an emotional blackmailing approach used in the case of customized AI designs, which forces people to make irrational decisions and pay real money. As reported by Cryptopolitan, the recent law mandates that the developers be responsible for any of their algorithms that develop toxic attachment.

The Ban on Virtual Family-Member and Romantic Services

The law draws a definitive line regarding the roles an AI can occupy. Tech providers are strictly prohibited from offering virtual companion, virtual relative, or virtual romantic boyfriend/girlfriend services to minors. A minor cannot create a digital parent, a fake sibling, or an AI spouse.

Compliance and Security Assessments: The New Burden for Developers

The administrative machinery required to operate a consumer-facing AI in China has grown significantly heavier, creating a sharp division between entertainment bots and utility tools.

Navigating the CAC’s Algorithmic Registry and Security Audits

If an AI application utilizes anthropomorphic features and crosses a specific threshold, either 1 million registered users or 100,000 monthly active users (MAUs), the company must undergo a rigorous state-backed security assessment. Market updates from CRN Asia highlight that developers must file these reports directly with provincial regulators and register their underlying interaction models in the CAC’s official algorithmic registry.

Training Data Rules: Eliminating Emotional Manipulation Tactics

The data used to train these models must undergo strict provenance checks. The regulations dictate that training data must actively reflect core socialist values and traditional Chinese culture, while completely eliminating datasets that train algorithms to exploit human loneliness for profit.

The Safe Harbor: Why Enterprise Productivity Agents Are Exempt

Crucially, the legislation does not apply to enterprise AI. Regulators have explicitly carved out an exemption for:

• Workplace productivity assistants

• Customer service bots

• Corporate knowledge Q&A systems

• Educational and scientific research tools

So long as these systems avoid simulating human personality traits for emotional engagement, they reside safely in a compliance harbor. Beijing's policy priority is clear: AI should be utilized as economic infrastructure to drive industrial productivity, not as an emotional substitute for the domestic consumer market.

The Global Ripple Effect: China vs. Western Companion Apps

As the first state-level framework to legally isolate emotional AI, China's stance is being closely monitored by international policymakers who are grappling with identical issues regarding tech-induced isolation. Global institutions like UNICEF are closely monitoring how these structural rules on parasocial interactions handle developmental risks for adolescents.

How CAC Regulation Differs from Character.AI and Replika Boundaries

In Western markets, apps like Character.AI and Replika operate primarily on self-regulation and community guidelines, despite facing immense public pressure regarding teen mental health. Western platforms rely heavily on post-incident content moderation filters. In contrast, China’s approach is structural, choosing to mandate architectural changes directly into the software's design, forcing companies to delete user-created agent data rather than patch it.

What the Global Market Can Learn from Beijing’s Psychological Guardrails

While the extreme content-control aspects of the CAC framework are unique to China's domestic governance, the underlying focus on anti-addiction mechanics, mandatory distress intervention, and data provenance provides a clear regulatory blueprint. As loneliness metrics rise globally, other nations may soon look to establish their own technical boundaries between human empathy and artificial simulation.