China Enforces Humanlike AI Rules, Agent Features Removed

For practitioners, this enforcement episode raises immediate questions about compliance-first product flags, dataset retention, and evaluation practices for systems that expose anthropomorphic behavior to end users. Companies operating conversational agents should treat rapid, region-specific rule changes as an operational risk that can force feature rollbacks and dataflows to change with little notice.

What happened - Reported facts: Reporting in Decrypt and in Amin Shamim's coverage states that major Chinese consumer apps have begun disabling persona/agent features ahead of the effective date of a new regulation. According to Amin Shamim, Doubao will take its agent function offline on July 15, 2026, while Amin Shamim and Decrypt report that Qwen disabled humanlike and user-created agents on July 10, 2026 and will disable broader agent services on July 15, 2026. Amin Shamim names the new rule as the "Interim Measures for the Administration of AI Anthropomorphic Interactive Services," and reports that it takes effect on July 15, 2026.

Industry-pattern observations: When regulators restrict 'anthropomorphic' behaviors, platforms typically implement conservative feature flags, limit user-created agents, and curtail persona-slot functionality to reduce legal exposure. That pattern increases the operational burden on testing and rollback tooling, and complicates A/B experimentation where user-facing behavior is the variable under test.

Context and significance - Editorial analysis: This is notable because the measures are framed specifically at "humanlike" AI interactions, a narrower target than general model governance. Observers following the sector will watch whether the rule's definitions produce a technical checklist (for example, banned response types, required disclaimers, or limits on simulated emotions) that other jurisdictions reference. Reporting so far focuses on the near-term product actions of ByteDance and Alibaba rather than detailed enforcement guidance.

What to watch - Editorial analysis: Track formal guidance or FAQs from Chinese regulators clarifying prohibited behaviors; monitor whether other platforms (Tencent, smaller app developers) follow with similar rollbacks; and watch for published compliance checklists that translate legal language into testable model behaviors. Also watch how operators handle archived conversational data and user-generated agent assets while disabling features, since data-retention and consent are common enforcement touchpoints.

Reported coverage comes from Decrypt and Amin Shamim's reporting. Neither source quoted a regulator's full text in our scraped excerpts; readers should consult the official measures for precise legal language.