Data poisoning and missing labels have always been hidden landmines. This Qinglang Action cleared 6 million pieces of non-compliant information in the first phase alone, with a severity stronger than expected. Compliance costs are likely to skyrocket going forward.

CoinWorld news: Since launching the "Qinglang·Rectifying AI Application Malpractices" special campaign in April 2026, the Cyberspace Administration of China has focused on issues such as failing to comply with large model registration obligations, inadequate AI platform security and review filtering capabilities, AI data poisoning, and insufficient implementation of generative content labeling. It has directed local cyberspace offices to advance the first phase of key rectification tasks and has set up a "Reporting Zone for AI Application Malpractices" to specifically handle public reports. Through joint efforts, the first phase has dealt with over 14k AI products, including illegal websites, applications, and agents; cleaned up over 14k pieces of illegal and harmful information; handled more than 26k accounts; removed over 1,300 illegal AI products and 9 illegal open-source datasets; and achieved positive progress in various tasks.