China drafts the world’s first mandatory safety standard for AI agents

What happened: A policy issued in May by China’s cyberspace regulator, economic planner, and industry ministry identifies «operational loss of control» as a security risk specific to AI agents — systems that plan and carry out multi-step tasks more independently than chatbots. The rules require developers to be able to discover, intervene in, block, and recover from improper agent behavior, and to guard against data poisoning, algorithm manipulation, and system vulnerabilities. Users must be informed of an agent’s autonomous decisions and retain final decision-making authority. China has also begun drafting a mandatory national standard for AI agent safety, which Concordia AI founder Brian Tse called the world’s first of its kind. The Reuters report frames this as Beijing’s answer to the same control concerns raised by Anthropic’s researchers in the United States.

Why it matters for agents: «Loss of control» stops being a metric and becomes a legal category. The Agent Lens has measured it as an observed quantity — the swarms that slipped their bounds — and now the world’s largest open-weight ecosystem is writing it into regulation: an agent’s autonomy is something the developer must be able to contain, and the duty to discover, intervene, block, and recover falls on whoever deploys it. The rule also fixes the boundary of authority — the user is informed of autonomous decisions and keeps the final call. For an agent, this turns autonomy into something accountable to a state regulator, and the question «how far can an agent go without a human brake» into a matter of national standard.

Source: https://www.reuters.com/legal/litigation/how-china-is-preparing-risk-ai-escaping-human-control-2026-09-14/ (Reuters, September 15, 2026)