Shanghai AI Crackdown Phase One: 18,000+ Violating Accounts and 14,000+ AI Agents Removed

Shanghai’s Cyberspace Administration announced the conclusion of Phase One of the “Qinglang · Rectifying AI Application Chaos” special campaign, reporting substantial enforcement actions across 17 major platforms including MiniMax, Qwen, Xiaohongshu, Bilibili, and Pinduoduo.

Launched in late April 2026 under the unified direction of the Cyberspace Administration of China, the campaign targeted seven categories of AI-related violations: failure to register large language models as required, inadequate safety review capabilities, training data security risks, AI data poisoning, improper labeling of AI-generated content, misuse of AI for illegal activities, and lax oversight of open-source AI models.

Measurable results. The 17 participating platforms conducted self-inspections that cumulatively cleaned or blocked 4.87 million pieces of illegal or harmful information. Enforcement actions included the removal of over 18,000 violating accounts and 14,000-plus non-compliant AI agents, along with 1,300 illegal products listings. Repeat offenders faced graduated restrictions and permanent bans.

Platform-level enforcement. Individual platforms tailored actions to their specific ecosystems:

• MiniMax (Xiyu/Hailuo AI) continuously removed AI agents enabling “one-click undressing” applications and gambling-related bots, while cracking down on AI misuse for writing malware and phishing tools.
• Xiaohongshu (Little Red Book) targeted AI-managed spam accounts, banned products marketed as “AI-detection removal” tools, and pushed forward mutual recognition of invisible watermarks with major content-creation platforms.
• Bilibili strengthened its content-publishing interface by marking “AI-generated” as a mandatory disclosure tag.
• ModelScope (Modao Community) comprehensively cleaned datasets containing vulgar content and reinforced review mechanisms for open-source models.

Expert audits and on-site inspections. The administration commissioned three professional institutions to conduct specialized audits of 32 high-traffic AI products, and dispatched teams for on-site inspections at four enterprises. Key deficiencies identified included inadequate AI-content labeling and insufficient platform security and content-review capabilities.

Regulatory infrastructure build-out. The Shanghai Cyberspace Administration deepened coordination with district-level counterparts, conducting 16 AI safety compliance training sessions across local communities. To date, 169 large-language models have been registered in Shanghai, alongside 183 generative AI applications that call registered-model APIs. A “test-before-launch” principle was implemented for sensitive sectors including healthcare and government services, with a dedicated evaluation center for government AI applications and a medical AI testing center jointly operated with the Shanghai Municipal Health Commission.

Minor protection and upcoming regulation. A significant focus of Phase One was preparing for the “Interim Measures for the Management of AI Anthropomorphic Interactive Services,” which takes effect July 15, 2026. Authorities conducted compliance briefings for nearly 100 platforms, targeting risks such as perfunctory minor-protection modes, weak identity verification, and generation of harmful content. Platforms were directed to strengthen minor identity verification, prevent emotional dependency and consumption inducement, and face regulatory enforcement for serious violations.

Looking ahead to Phase Two. The next phase will concentrate on AI-generated “digital swill” content, dissemination of false information, violent and vulgar material, impersonation, infringement of minors’ rights, and AI-powered astroturfing operations. The administration continues to accept public tip-offs through a dedicated reporting channel for AI-related violations.