Summary
On July 15, 2026, China's Implementation Opinions on AI agents became enforceable — the world's first regulatory category dedicated to AI agents. The framework sorts agent decision authority into three tiers (human-only, approval-required, autonomous), mandates filing and compliance testing in sensitive sectors, and requires that users retain final decision-making power with human override.
What changed
China's Implementation Opinions on AI agents became enforceable on July 15, 2026, establishing a three-tier decision-authorization framework (human-only, approval-required, autonomous), mandatory filing and compliance testing for agents in healthcare, finance, transportation, and public safety, plus human-override and product-recall provisions.
Why it matters
As the first binding, agent-specific regime, it forces any operator in the Chinese market to document and enforce autonomy limits before deployment, turning agent governance from best practice into a legal requirement. It sets an early template that enterprise agent platforms and regulators elsewhere will be measured against.
Evidence excerpt
China's Implementation Opinions on AI agents became enforceable July 15, 2026 — the world's first dedicated AI-agent regulatory category — sorting decision authority into three tiers (human-only, approval-required, autonomous) with mandatory filing and human override in sensitive sectors.