scmp.com web signal

ByteDance and Alibaba disable humanlike AI agents before July 15

3 sources tracking this story
bytedance alibaba agents china ai regulation ai-business china regulation

TL;DR

  • Five central agencies co-issued the rules, signaling state consensus rather than a single regulator's initiative and reducing the likelihood of quiet reversal.
  • Anti-addiction friction requirements are structurally incompatible with persistent-memory agent design, making reinstatement of companion features operationally costly.
  • ByteDance's October 15 data deletion mandate means any reinstated Doubao companion service would restart user relationships from scratch.

China's incoming rulebook for consumer AI is about to switch on, and the two biggest domestic platforms are moving to comply days ahead of the switch. According to the South China Morning Post, ByteDance's Doubao notified users on July 5 that its agent feature will go offline on July 15, citing "product function adjustments," with user data becoming inaccessible after October 15. Alibaba's Qwen posted a similar notice on July 6, saying its humanlike interactive agents and user-created agent functions will be disabled on July 10, and broader agent services on July 15.

The trigger is the Interim Measures for the Administration of Artificial Intelligence Anthropomorphic Interaction Services, effective July 15. The framework covers systems that "simulate human personality traits, thinking patterns and communication styles to provide sustained emotional" interaction, and cites risks around extremist ideas, privacy leaks, harm to physical and mental health, and dependence or addiction. Customer service bots, knowledge Q&A, workplace assistants, and education and scientific research tools are carved out, as long as they do not involve sustained emotional interaction.

That carve-out is doing a lot of work. The SCMP piece notes that both platforms previously let users create customized agents with specific personas, including tutors, role-playing characters, and companions with fixed communication styles, and it is exactly that category the rules target. What is being pulled is not a small feature. It is the whole surface that let users bake in a persona and keep coming back to it.

The honest caveat is that the reporting does not spell out enforcement, including penalties, how the sustained-emotional-interaction line gets adjudicated in edge cases, or whether the ban touches enterprise deployments as well as consumer apps. It also does not say whether smaller Chinese chatbot apps have followed suit, though the fact that both flagships are complying rather than pushing on the definitions is itself a signal about how the regulator's line is being drawn.

For anyone building on Chinese AI platforms, the workplace-assistant framing is now the safer design surface. Q&A, customer service, and productivity assistants sit inside the exemption, while anything that reads as a persona or a companion sits outside it. Expect Doubao and Qwen to redirect product effort accordingly.

What others are reporting

Coverage cluster as of 24h after publish

  1. The Next Web Read →

    Adds expert framing from Pan Helin on the policy rationale and connects the shutdowns to Beijing's parallel investment vetting plans, situating compliance within a broader sectoral control framework.

    "Current agents are not yet mature," Pan Helin said, explaining the policy prioritizes "safety, practical use, and standardisation."
  2. AI News Read →

    Details the specific compliance machinery -- eight-area security assessments and provincial filing requirements -- and maps the implementation gaps around technical thresholds, liability, and data portability.

    "The rules draw a line between the agent that does your work and the agent that keeps you company, and it is only the second kind that Beijing has moved against."