中/EN
Frontline Hotspot
Frontline Hotspot

Doubao and Qianwen Shut Down Agents Same Day: Impact of China's AI Personification Regulation

ByteDance's Doubao and Alibaba's Qianwen simultaneously shut down user-built AI character features due to the "Interim Measures for AI Personified Interactive Services" taking effect July 15. Platforms cannot comprehensively review massive UGC agents, resulting in blanket shutdowns. User digital asset property rights issues highlighted.

Published July 27, 20264 min read

1. The Event: Doubao and Qianwen "Kill" Agents on the Same Day

On July 4, ByteDance's Doubao and Alibaba's Tongyi Qianwen simultaneously announced:

User-built AI characters, persona dialogues, workflows, and other agent features would be officially shut down on July 15.

Doubao offered a viewing buffer period until October 15; Qianwen would become inaccessible immediately after shutdown. Users needed to back up configurations and conversation history via screenshots or text export before July 15.

Millions of users' "digital avatars," cultivated over months, were wiped out overnight.

2. The Cause: "Interim Measures for AI Personified Interactive Services"

This shutdown wasn't "voluntary" by the platforms—it was compliance pressure.

On July 15, 2026, the "Interim Measures for the Management of AI Personified Interactive Services," jointly issued by five departments, officially took effect. The measures impose compliance review requirements on personified interactive services:

  • Providing personified interactive services requires appropriate qualifications
  • AI characters must pass content safety reviews
  • User-generated content (UGC) requires platforms to bear primary responsibility

The problem: platforms cannot conduct comprehensive reviews of massive user-built agents.

A platform might have millions of user-created characters, each with different personas, dialogue styles, and knowledge boundaries. Review them one by one? Prohibitively expensive. Don't review? Non-compliant.

The platforms' rational choice: blanket shutdown.

3. The Impact: Who Pays for "Digital Memories"?

This incident exposes a deeper question: Who owns the AI characters users create?

Users invest time, effort, and emotion into training an AI character—writing personas, feeding data, correcting tone. To the platform, this character is just a database record; to the user, it might be a "digital friend" or even an "emotional anchor."

But under current legal frameworks:

  • Platform user agreements typically state "the platform reserves the right to terminate services at any time"
  • The legal nature of user-created AI character configurations is unclear (creative work? data? virtual property?)
  • Platforms can shut down without compensation; users have no recourse

This isn't a technical problem—it's a property rights problem.

4. Industry Chain Reactions

Doubao and Qianwen aren't isolated cases. Expect:

  1. Smaller platforms struggle more: Big companies can invest in compliance teams for reviews; smaller platforms may simply abandon UGC agent features.
  2. Open-source solutions benefit: Locally deployed open-source models (Ollama + Llama 4, Qwen3, etc.) aren't subject to these restrictions. "Raise your own AI" becomes a new trend.
  3. Enterprise market expands: Enterprise-built AI assistants (with clear responsible parties) actually accelerate, because compliance pathways are clear.

5. Advice for Ordinary Users

  1. Back up important configurations: UGC content on any platform can go offline due to policy changes. Develop a habit of regular exports.
  2. Consider local solutions: If you deeply depend on AI characters, local deployment (Ollama + open-source models) is the only "unpluggable" option.
  3. Reduce emotional dependence: AI characters are tools, not friends. Platforms can shut them down anytime; moderate emotional investment is wise.

In one sentence: The Interim Measures' intentions are good (preventing chaos in AI personified services), but the "blanket shutdown" execution wiped out millions of users' digital assets instantly. Regulation needs more refined tiered management, not "can't manage it, so shut it down."

References:

This article is AI-assisted and human-edited. Last updated: 2026-07-27

FAQ

Why did platforms choose blanket shutdowns instead of individual reviews?
Because the cost is prohibitive. A platform might have millions of user-created characters, each with different personas, dialogue styles, and knowledge boundaries—individual review is economically unfeasible.
Do users have legal protection for AI characters they created?
Currently the legal nature is unclear. Platform user agreements typically state "the platform reserves the right to terminate services at any time." Whether user-created AI character configurations constitute creative works, data, or virtual property remains undetermined.
Can locally deployed open-source models circumvent this restriction?
Yes. The Interim Measures target platforms "providing personified interactive services." Individuals locally deploying open-source models (e.g., Ollama + Qwen3) are not service providers and are not subject to this restriction.

Related

Frontline Hotspot

One prompt to final cut: JianYing Hub closes the AI video loop

According to a 9-21 report by Qbit, ByteDance's JianYing launched JianYing Hub, a one-stop AI video creation entry point on PC, whose product move is not about model parameters but about workflow, welding generation and editing into a single entry. Official positioning is a PC-side one-stop AI video creation workbench; the official page lists nine core functions (AI image and asset generation, storyboard scripting, module wiring and asset management, batch storyboard prompt generation, multi-model video generation with preview, direct hand-off to editing, AI post-editing, the JianYing Assistant Agent, and ByteDance asset import) along with a 14-step onboarding path and an official comparison table against Jimeng AI (source-side framing, not independently retested here). Two real changes stand out: generation results are not exported and re-imported but jump straight via "More Editing" into JianYing's multi-track timeline for AI extend, upscaling, frame interpolation, color grading, removal and vocal separation, an in-project closed loop replacing file exchange; and the JianYing Assistant Agent turns repetitive work into a single sentence by calling Skills for cutting voiceover, adding narration, fixing subtitles and batch production. The article's own judgment is that a workbench solves the last mile from asset to publishable cut rather than the ceiling of image quality, and that Hub is an orchestration layer rather than a generation engine, with three costs of the loop, ecosystem lock-in, tight asset-and-account coupling, and opaque pricing. Pricing, free quota, concurrency, credit rules, regional availability and duration or resolution limits are all unpublished and are stated as following the official app, with no invented numbers, and the launch timing is only a second-hand report.

Sep 22, 20267 min read
Frontline Hotspot

From 2.8s to 2.3s: can Qwen3.8 steal the interpreter's job?

In September 2026 Alibaba's Qwen team released Qwen3.8-LiveTranslate, a real-time simultaneous interpretation model opened through the Qwen AI platform and Alibaba Cloud Bailian as a WebSocket streaming API that can be embedded in meeting systems, live streams and support desks. Headline figures: average lag (LAAL) cut from 2.8 to 2.3 seconds; recognition input in 60 languages and speech output in 29; three capabilities, real-time speaker diarization plus voice cloning, source and translation emitted in the same frame, and long-context disambiguation, with video and audio input helping resolve ambiguity. Technically it rests on an Interleave single-stream architecture that caches already-heard audio and already-emitted translation instead of reprocessing each sentence, plus a Hybrid MoE Thinker-Talker pair, where the Thinker arranges video, audio, source and translation into one causal sequence and the Talker fuses translation with source audio into speech that keeps the original speaker's timbre. The article keeps its figures honest: 2.3 seconds is average lag rather than end-to-end first-packet latency, 60 and 29 are different units, the vendor comparison table is not independently retested, an unpublished metric is not the same as a bad one, pricing, rate limits, concurrency and regional availability are not invented, and the model is an API service rather than open source.

Sep 21, 20267 min read