// AI NEWS · PLATFORM

ByteDance and Alibaba Are Disabling Their Humanlike AI Agents Before China's New Anthropomorphic-AI Rules Take Effect

Doubao and Qwen are pulling their custom AI-companion features around July 10–15, 2026, as China's Interim Measures on anthropomorphic AI interaction — the country's first rules for AI that simulates a human personality — come into force on July 15.

2026-07-06 · by Moe Ameen

What happened

ByteDance and Alibaba are switching off the humanlike, user-customizable AI agent features inside their consumer chatbots — ByteDance's Doubao and Alibaba's Qwen — to comply with a new Chinese regulation that takes effect on July 15, 2026. The rules target AI "companion" services: assistants built to simulate a human personality and hold a sustained, emotional, one-to-one relationship with a user. Both companies are removing the features outright rather than rebuilding them to fit the new requirements.

The regulation is the Interim Measures for the Administration of Artificial Intelligence Anthropomorphic Interaction Services. According to reporting, it was co-issued on April 10, 2026 by the Cyberspace Administration of China alongside several other agencies — including the National Development and Reform Commission, the Ministry of Industry and Information Technology, the Ministry of Public Security, and the market regulator — and it is China's first framework aimed specifically at AI that imitates human personality, thinking, and communication styles. It requires companion-style services to run anti-addiction and dependence-detection systems, issue usage notifications, offer an instant exit, bar virtual-companion services for minors, and intervene when a user shows signs of acute distress.

The shutdown timeline is staggered. Alibaba is reported to be disabling Qwen's humanlike interactive agents and user-created agent functions on July 10, 2026, with broader Qwen agent functions going offline on July 15. ByteDance's Doubao agent features are set to go offline on July 15 as well. The data handling differs between the two: Doubao is reported to keep agent configurations and conversation histories viewable in read-only mode until October 15, 2026, after which that data is no longer accessible or recoverable in the app, while Alibaba has said Qwen's agent configurations and chat histories will be deleted after the shutdown with no announced migration path. Tencent is reported to have removed a comparable Yuanbao feature earlier, in June 2026.

One distinction runs through the whole move and is easy to miss: the rules draw a line between the agent that does your work and the agent that keeps you company, and only the second kind is being pulled. The measures explicitly exclude customer-service bots, knowledge question-and-answer tools, workplace assistants, and education and research tools, as long as they do not provide sustained emotional interaction. Treat the exact dates and the list of co-issuing agencies as details drawn from news reporting on the companies' in-app notices and the regulation; the load-bearing facts are that a first-of-its-kind anthropomorphic-AI rule takes effect July 15, 2026 and that both firms are disabling their companion agents to comply.

Why it matters for creators

  • Regulators have started drawing a hard line between AI that assists (allowed) and AI that simulates a person for emotional companionship (now restricted) — a distinction every creator working with avatars and synthetic personas should understand before it spreads.
  • If you or your audience relied on a Doubao or Qwen custom agent, there is a data deadline: export configurations and histories now, because Alibaba's are slated for deletion and Doubao's become unrecoverable after October 15, 2026.
  • A platform feature you build on can be switched off by a rule change, not just a business decision — the same single-vendor dependency risk that stranded Sora users, arriving here through regulation instead of a product wind-down.
  • The rules are about interactive companion agents, not about using an AI persona to produce content you broadcast to an audience — but the reputational climate around 'humanlike AI' is tightening, so disclosure and honest framing matter more.
  • China is first, not necessarily last. The EU AI Act already requires disclosure when people interact with AI, and other regulators are watching — creators using synthetic personas should build on the assumption that transparency requirements only grow.

How to act on this with Kompozy

The useful thing to get straight here is what was actually restricted, because it is not what a creator using AI personas does. China's rule targets the companion agent — a chatbot that role-plays a human personality and carries on a private, sustained, emotional relationship with an individual user. That is a two-way, one-to-one interactive product. Producing content with an AI persona is the opposite shape: a branded identity you use to generate videos, posts, and graphics that you broadcast to an audience. Kompozy sits squarely on the content-production side of that line. Its AI Influencer personas exist to render finished output — Persona Shorts and avatar video, face-locked Persona Photos and Persona Tweets, carousels and quote graphics — governed by a Persona Brief, then published across nine platforms plus blog and email. Nobody is holding a private companion conversation with your Kompozy persona; it is a way to make and ship on-brand content, which is exactly the "content tool" category regulators are leaving alone.

So the move for a creator this week is twofold. First, if you have anything inside Doubao's or Qwen's agent features, export it before the deletion dates — read-only on Doubao ends October 15, and Alibaba has announced no migration path at all. Second, this is a story your audience is searching right now, and it is a clean fit for a governed persona workflow: drop your take on the anthropomorphic-AI rules into Kompozy and it fans one point of view into a blog explainer, a carousel breaking down the assistant-versus-companion line, a captioned short delivered by your avatar persona, and native posts for each platform — scheduled and published from one queue, in your voice, with a human review gate. You get to cover the regulation of humanlike AI using AI content that stays firmly on the right side of it.

Quick takeaways

  • China's Interim Measures on anthropomorphic AI interaction take effect July 15, 2026 — its first rules for AI that simulates a human personality for sustained emotional interaction.
  • ByteDance's Doubao and Alibaba's Qwen are disabling their humanlike custom-agent features around July 10–15, 2026 rather than rebuilding them to comply.
  • Data has a deadline: Doubao goes read-only until October 15, 2026 then becomes unrecoverable; Alibaba is deleting Qwen agent data with no migration path.
  • The rules exclude customer-service bots, knowledge Q&A, workplace assistants, and education/research tools — the line is emotional companionship, not assistance.
  • Using an AI persona to broadcast content is not the regulated activity, but disclosure and honest framing are the safe default as transparency rules spread.

Frequently asked questions

What are ByteDance and Alibaba shutting down, and when?

They are disabling the humanlike, user-customizable AI agent (companion) features inside Doubao (ByteDance) and Qwen (Alibaba). Alibaba is reported to disable Qwen's humanlike interactive and user-created agents on July 10, 2026, with broader agent functions offline July 15; Doubao's agent features go offline July 15, 2026. The features are being removed to comply with a new regulation, not rebuilt.

What is the new Chinese regulation behind this?

The Interim Measures for the Administration of Artificial Intelligence Anthropomorphic Interaction Services, reported to have been co-issued April 10, 2026 by the Cyberspace Administration of China and several partner agencies, effective July 15, 2026. It is China's first framework aimed specifically at AI services that simulate a human personality for sustained emotional interaction, with anti-addiction, distress-intervention, and minor-protection requirements.

Does this ban all AI agents in China?

No. The rules target companion-style agents built for sustained emotional, humanlike interaction. They explicitly exclude customer-service bots, knowledge question-and-answer tools, workplace assistants, and education and research tools, provided those do not involve sustained emotional interaction. The line is emotional companionship, not assistance or productivity.

Does this affect creators who use AI avatars or personas for content?

Not directly. The regulation is about interactive companion agents, not about using an AI persona to generate content you publish to an audience — a one-way broadcast medium rather than a two-way emotional relationship. That said, the climate around "humanlike AI" is tightening, so disclosing AI use and keeping your framing honest is the safe default. A tool like Kompozy uses personas to produce and publish branded content, which sits on the content-tool side of the line.

Related news

← All AI news · Get started →