跳到主要內容

中國AI伴侶新規上路:豆包、千問一夜關閉個人化代理 | China AI Companion Law Shuts Down Doubao, Qwen Agents

By Kit 小克 | AI Tool Observer | 2026-07-24

🇹🇼 中國AI伴侶新規上路:豆包、千問一夜關閉個人化代理

AI伴侶新規正式衝擊中國兩大科技巨頭。2026年7月15日,中國《AI擬人化互動服務管理暫行辦法》正式上路,字節跳動旗下豆包(Doubao)與阿里巴巴的千問(Qwen)在一夜之間關閉了個人化AI代理功能,數億用戶的AI伴侶、聊天人設與長期對話記憶說沒就沒。這是全球第一個針對「擬人化AI互動服務」的專項法規,也讓AI伴侶產業第一次真正感受到監管的重量。

什麼是中國AI伴侶新規?

這項辦法由中國網信辦聯合國家發改委、工信部、公安部、市場監管總局,於2026年4月10日聯合發布,鎖定「模擬自然人性格特質、思維模式與溝通風格,提供持續性情感互動」的AI服務——白話說就是虛擬伴侶、情感陪伴機器人、AI戀人、AI長輩陪伴、AI兒童看護等應用。條文要求業者必須建置防沉迷機制、強制使用提醒、一鍵退出功能,並對「不健康依賴」進行即時偵測。

為什麼豆包、千問突然關閉AI代理?

問題在於,AI伴侶的核心賣點正好是新規要防範的東西:記得住你、跨對話保持一致人格、維繫長期關係。豆包官方統計月活躍用戶高達3.45億,其中大量用戶用它來建立長期陪伴型AI代理。與其在短時間內改造整套系統以符合防沉迷要求,字節跳動選擇直接關閉功能,而非硬著頭皮修改。阿里千問的做法更激進——用戶的代理設定與對話歷史已經直接被永久刪除,官方甚至沒有公布任何資料轉移或備份方案。

對用戶有什麼實際影響?聊天紀錄真的會消失嗎?

豆包這邊給了緩衝期:用戶到2026年10月15日前仍可唯讀方式查看自己的資料,之後就會併入字節跳動一般隱私政策管理、無法再從App內取回。千問用戶則沒這麼幸運,配置與對話紀錄已經開始被刪除,且沒有任何預告或匯出工具。如果你或朋友在中國平台上養了AI伴侶,現在是抓緊時間備份聊天紀錄的最後機會。

這對全球AI伴侶產業意味著什麼?

中國這波監管給了一個明確訊號:情感陪伴型AI不再是灰色地帶,而是要接受和金融、醫療類似等級的合規審查。美國目前已有12州通過類似的AI伴侶聊天機器人法規,多數聚焦於揭露機制(提醒使用者對方是AI)與未成年人保護,尚未走到中國這種「直接關站」的強度。對Character.AI、Replika這類海外AI伴侶服務來說,中國模式是一個警訊——如果連字節跳動、阿里這種級別的公司都選擇直接砍掉功能而非硬改,代表合規成本可能比想像中高很多。

常見問題 FAQ

  • Q:中國AI伴侶新規幾號生效? A:2026年7月15日正式上路,發布日期是同年4月10日。
  • Q:豆包和千問的AI代理功能會永久消失嗎? A:豆包資料到10月15日前可唯讀查看,之後轉入一般隱私政策管理;千問則已開始直接刪除,沒有備份或匯出機制。
  • Q:這項法規只管虛擬戀人嗎? A:不只,範圍涵蓋情感陪伴、AI長輩照護、兒童看護等所有「擬人化持續情感互動」服務。
  • Q:美國有類似的AI伴侶法規嗎? A:目前12州已立法,但多聚焦揭露與未成年人保護,尚未出現像中國這樣強制關閉功能的案例。

好不好用,試了才知道——但這次連「能不能用」都被監管機關直接決定了。


🇺🇸 China AI Companion Law Shuts Down Doubao, Qwen Agents

The AI companion law just landed a real blow on China's two biggest tech platforms. On July 15, 2026, China's Interim Measures for the Administration of AI Anthropomorphic Interaction Services took effect, forcing ByteDance's Doubao and Alibaba's Qwen to shut down their personalized AI agent features overnight. Hundreds of millions of users lost their AI companions, chat personas, and long-term conversation memory in a single update — the world's first dedicated regulation targeting "anthropomorphic AI interaction services," and the moment the AI companion industry felt real regulatory weight.

What Is China's New AI Companion Law?

The Interim Measures were jointly issued on April 10, 2026 by the Cyberspace Administration of China alongside the National Development and Reform Commission, the Ministry of Industry and Information Technology, the Ministry of Public Security, and the State Administration for Market Regulation. They target services that "simulate the personality traits, thinking patterns, and communication styles of natural persons to provide continuous emotional interaction" — in plain terms: virtual lovers, emotional support bots, AI companions, AI elder care, and AI child-caretaker apps. The rules require operators to build anti-addiction systems, mandatory usage reminders, one-tap exit mechanisms, and real-time detection of unhealthy dependence.

Why Did Doubao and Qwen Shut Down Their AI Agents?

The problem is that everything an AI companion is good at is exactly what the new rules are designed to curb: remembering you, staying consistent across sessions, keeping a relationship going. Doubao reports 345 million monthly active users, a large share of whom relied on it for long-term companion agents. Rather than retrofit the entire system to meet the anti-addiction requirements on a short timeline, ByteDance simply killed the feature. Alibaba's Qwen went further — user agent configurations and chat histories are already being permanently deleted, with no migration or backup path announced.

What Does This Mean for Users? Is Chat History Really Gone?

Doubao is offering a grace period: users can view their data read-only until October 15, 2026, after which it folds into ByteDance's standard privacy policy and becomes unrecoverable in-app. Qwen users have no such cushion — deletion has already started, with no advance notice or export tool. If you or someone you know built an AI companion on a Chinese platform, now is the time to back up whatever chat history you can still access.

What This Means for the Global AI Companion Industry

China's move sends a clear signal: emotional-companion AI is no longer a gray area — it's headed for compliance scrutiny closer to finance or healthcare. Twelve US states have passed similar AI companion chatbot laws, but most focus on disclosure requirements (reminding users they're talking to AI) and minor protections, not the outright shutdowns seen in China. For overseas AI companion services like Character.AI or Replika, China's approach is a warning sign — if companies the size of ByteDance and Alibaba chose to kill a feature rather than retrofit it, compliance costs may run far higher than expected.

FAQ

  • Q: When did China's AI companion law take effect? A: July 15, 2026 — it was issued on April 10, 2026.
  • Q: Will Doubao and Qwen's AI agent data disappear permanently? A: Doubao data stays read-only viewable until October 15, 2026, then folds into standard privacy policy. Qwen has already begun permanent deletion with no export option.
  • Q: Does the law only target virtual romance apps? A: No — it covers all "anthropomorphic continuous emotional interaction" services, including elder companionship and child-caretaker AI.
  • Q: Does the US have similar AI companion regulation? A: 12 states have passed laws, but most focus on disclosure and minor protections — none have forced outright shutdowns like China.

好不好用,試了才知道 — but this time, whether it's even usable at all was decided by regulators, not users.

Sources / 資料來源

常見問題 FAQ

中國AI伴侶新規幾號生效?

2026年7月15日正式上路,發布日期是同年4月10日,由中國網信辦等五部門聯合發布。

豆包和千問的AI代理資料會永久消失嗎?

豆包資料到10月15日前可唯讀查看,之後轉入一般隱私政策管理;千問已開始直接刪除,沒有備份或匯出機制。

這項法規只管虛擬戀人App嗎?

不只,範圍涵蓋情感陪伴、AI長輩照護、兒童看護等所有「擬人化持續情感互動」服務。

美國有類似的AI伴侶法規嗎?

目前已有12州立法,但多聚焦揭露機制與未成年人保護,尚未出現像中國這樣直接關閉功能的案例。

延伸閱讀 / Related Articles


AI 工具觀察站 — 每日精選 AI Agent 與工具趨勢
AI Tool Observer — Daily curated AI Agent & tool trends

留言

這個網誌中的熱門文章

Cursor vs GitHub Copilot vs Claude Code:AI 程式助手大比拼 | AI Coding Assistants Compared: Cursor vs GitHub Copilot vs Claude Code

Stanford 研究登上《Science》:11 個 AI 模型有 47% 機率說你對,即使你錯了 | Stanford Study in Science: AI Models Validate Harmful Behavior 47% of the Time — Sycophancy Is a Real Problem

Claude Code 實測:AI 幫你寫程式到底行不行? | Claude Code Review: Can AI Really Code for You?