China's Anthropomorphic AI Law Takes Effect Today, Wiping Millions of User-Built Agents from Doubao and Qwen
China's Interim Measures for the Administration of AI Anthropomorphic Interactive Services took effect today, July 15, 2026, making Beijing the first government on earth to impose a dedicated regulatory framework specifically targeting AI systems designed to simulate human personality and sustained emotional interaction. The rule was co-issued in April by five agencies — the Cyberspace Administration of China, 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 — and has been sitting in a grace period ever since. The enforcement warm-up, as reporting by Artiverse and TechNode documented, was not gentle: Shanghai authorities conducted a pre-deadline sweep that removed more than 14,000 non-compliant AI agents from platforms including Xiaohongshu and Bilibili, targeting impersonation bots, gambling tools, and applications generating prohibited imagery on request.
The practical consequence for China's two largest consumer AI platforms has been severe and swift. ByteDance's Doubao and Alibaba's Qwen both announced the discontinuation of their user-created agent features effective today. As reported by TechNode and the South China Morning Post, after the shutdown users will no longer be able to create new AI agents, and all existing user-built agents will also stop functioning. Doubao is permitting users read-only access to their agent configurations and chat histories until October 15, after which the data will be processed under its privacy policy and become unrecoverable. Qwen has offered no equivalent grace period, with agent configurations and conversation histories set for permanent deletion. Tencent's Yuanbao had already moved earlier, pulling a comparable feature in late June, meaning all three of China's dominant consumer AI chatbots have now exited the companion-agent market in a coordinated retreat.
The regulation draws a deliberate line between productivity agents — customer service bots, research tools, enterprise workflow assistants — which remain fully encouraged, and companion agents designed for sustained emotional engagement and persistent memory, which are now the primary target. The measures require companion services to run anti-addiction systems, issue mandatory usage notifications, offer instant-exit mechanisms, and detect signs of unhealthy dependence in real time, as Artificial Intelligence News explained. Those obligations sit in fundamental architectural conflict with agents built to maintain continuity across sessions, and rather than rebuild their products from scratch to meet compliance demands, ByteDance and Alibaba chose to shut the features down entirely. ByteDance is redirecting Doubao users to Maoxiang, a separate application purpose-built for regulatory compliance; Alibaba has announced no equivalent migration path for Qwen users. The user backlash on Weibo has been visible and angry, with users describing companion agents as long-standing emotional support systems whose loss leaves no easy alternative.
The economic context adds a layer of complexity that the regulatory framing obscures. As Robo Rhythms reported, citing Chinese financial press, Doubao's companion features had driven the platform to 180 trillion tokens of daily usage by late June, a staggering growth of roughly 1,500 times since May 2024 — yet the feature generated minimal direct revenue against an enormous compute bill. ByteDance's decision to shut down rather than retrofit is therefore not purely a capitulation to regulators; it may also reflect a rational business calculation that the cost of compliance engineering exceeded the economic return of a feature that was effectively free at scale. Alibaba's calculus appears similar. What Beijing has achieved, intentionally or otherwise, is the elimination of a product category that was economically marginal for platforms but deeply meaningful for users, a gap that now creates a significant opportunity for any compliant successor product.
The global implications of China's move are already being tracked by regulators in other jurisdictions. As Tech Times noted, California's SB 243 entered force in January 2026 as the first US state law to regulate AI companion services, while European institutions are advancing separate frameworks targeting addictive digital systems and AI emotional engagement. The FTC separately opened a public comment period this month on a proposed policy statement addressing state-level AI accuracy laws, a process running until July 31, reflecting the broader US struggle to balance federal consistency against state-level experimentation. Beijing's willingness to impose hard product requirements — rather than principles-based guidance — on two of its most globally competitive AI companies offers a template that other governments may be watching more carefully than they publicly admit. Whether that template leads to safer AI ecosystems or simply pushes emotional-AI development into less regulated corners of the global market is a question today's shutdowns cannot yet answer.