China’s AI Purge Hits 14K Products, 26K Accounts as Bitcoin Holds $64K

China’s Cyberspace Administration (CAC) has removed more than 14,000 non-compliant artificial intelligence products from domestic networks in the opening phase of its “Qinglang” enforcement drive, according to an official filing reviewed for this report. The purge, which covers websites, apps and autonomous AI agents, also wiped over 6 million pieces of illegal or harmful content from local platforms. For desks tracking AI-linked crypto tooling — trading bots, AI-driven wallets, agentic execution layers — the sweep marks the broadest state intervention into AI infrastructure to date, with direct implications for compliance risk on anything built atop Chinese-hosted AI stacks.

26,000 accounts suspended, nine datasets banned

Regulators went beyond product takedowns. The filing states more than 26,000 accounts were suspended and over 1,300 AI product listings were pulled from marketplaces. Nine open-source datasets were also declared illegal under existing Chinese rules — a move that redefines, in practical terms, what domestic developers can legally train on going forward.

The campaign, which began in April 2026, targets four structural gaps: bypassed mandatory model registration, weak safety filtering, deliberate data poisoning, and unlabeled machine-generated content. Data poisoning — corrupting training sets to manipulate model output — drew particular scrutiny given its potential to distort automated decision systems at scale, a category that includes AI-driven trading infrastructure increasingly deployed across crypto markets.

Big tech races to comply; some firms cut features instead

Alibaba has upgraded content-identification systems, while Huawei added review layers inside its app store to flag non-compliant listings before publication. Zhipu built a new review model, and DeepSeek deployed verification checks designed to intercept data manipulation earlier in the pipeline. The speed of the response signals how seriously the largest players are treating enforcement risk, a posture that contrasts with the more adversarial regulatory friction seen in Western markets over the past two years.

Other firms took a defensive route. ByteDance’s Doubao and the Qwen team disabled custom agent features entirely rather than risk falling short on registration and labeling requirements. Cutting functionality outright, instead of retrofitting compliance, underscores how costly meeting the four-point standard has become in practice — a signal worth noting for any team building AI crypto wallets or agent-based execution tools on Chinese infrastructure, where unregistered deployment now carries existential risk rather than a warning.

Market backdrop stays defensive

The regulatory tightening lands against a cautious tape. Bitcoin is trading near $64,000, the Fear and Greed Index sits at 24, Bitcoin dominance has climbed to 69.4%, and total crypto market capitalization stands at roughly $1.85 trillion. With risk appetite already compressed, added compliance overhead for AI-linked crypto tooling in one of the largest tech jurisdictions adds another variable for desks pricing regulatory tail risk into 2026.

Read more: Bitcoin Pinned Near $64K as Fear Index Hits 24, Dominance Climbs to 69.3%