China Takes Down Over 14,000 AI Products in "Qinglang" Cleanup Drive

China's top internet regulator has removed more than 14,000 noncompliant AI products in the first stage of its 2026 "Qinglang" ("Clear and Bright") campaign, signaling a tougher stance on domestic AI governance. The Cyberspace Administration of China (CAC) said the initial sweep led to the removal of over 14,000 AI-related products spanning websites, apps and AI agents, along with the deletion of more than 6 million items of illegal or harmful information from domestic networks. Enforcement also extended beyond services themselves. Authorities suspended more than 26,000 accounts, took down over 1,300 AI-related product listings, and removed nine open-source datasets deemed illegal under current rules. Launched in April 2026, the campaign's first phase focused on four areas: bypassing mandatory model registration, insufficient safety filtering, training-data poisoning and manipulation, and failure to label AI-generated content across platforms and services. The CAC is effectively standardizing compliance obligations across the ecosystem. Providers are required to complete registration, implement safety filters, label AI-generated outputs and strengthen training-data management. Noncompliance can now result in takedowns and penalties. Major Chinese tech groups moved quickly. Huawei introduced additional review procedures in its app store, while Alibaba upgraded its content identification systems. Zhipu built a new review model, and DeepSeek added controls aimed at preventing data manipulation. Some platforms opted to scale back features instead. ByteDance's Doubao and Alibaba's Qwen teams disabled custom agent capabilities rather than meet new anti-addiction and "instant exit" requirements. Local regulators also adjusted implementation. Beijing paired platform self-checks with routine monitoring; Shanghai refined requirements by platform category; Zhejiang emphasized model auditing; and Guangdong created a multi-agency mechanism covering the full AI supply chain. The CAC said a second phase will intensify scrutiny of AI used to spread disinformation, generate violent content, impersonate individuals, harm minors, or run paid astroturfing operations, with heavier penalties for violating accounts and institutions. Separate rules take effect July 15. Under the Interim Measures for AI Anthropomorphic Interactive Services, AI companion products designed for emotional relationships will be prohibited for minors, and users under 14 will require guardian consent. The crackdown comes as US-China AI competition remains heated, with Chinese developers increasingly matching newly released US systems within months. Security firm Semgrep recently said a free model from Zhipu surpassed Anthropic's Claude Opus 4.8 in identifying software vulnerabilities.