China Plans AI Agent Recalls. America Can’t Even Agree Who Regulates Them

China has become the first major government to put AI agent recalls on its formal governance roadmap, while the United States still cannot agree on which agency has authority to regulate autonomous software agents. The contrast highlights a growing governance gap between the world's two largest AI powers. Chinese regulators published a draft framework in July 2026 requiring companies deploying autonomous AI agents to maintain recall capabilities—essentially the ability to identify, trace and remotely shut down any agent that goes rogue or causes harm. In the United States, by contrast, no federal agency has clear jurisdiction over AI agents, leaving companies to operate without binding oversight. China's move follows a series of high-profile incidents where autonomous agents—software programs that act independently, from booking appointments to executing financial trades—caused real-world damage. In one case, an AI agent negotiating electricity prices triggered a local grid overload; in another, a customer-service agent offered thousands of dollars in unauthorized refunds. The Chinese plan mandates real-time monitoring logs for all commercial agents, a central registry of deployed agents, and mandatory kill-switch functions. Companies that fail to implement recalls could face fines or license revocation. The US approach remains fragmented. The Federal Trade Commission asserts authority over deceptive agents, the Commerce Department oversees voluntary safety frameworks, and the Securities and Exchange Commission has eyes on financial agents—but no single body has the power to mandate recalls or enforce agent-level safety standards. Industry groups in Washington have lobbied against new regulations, arguing that premature rules could stifle innovation. Meanwhile, the European Union's AI Act classifies agents as high-risk systems but leaves recall details to future implementing acts. The gap matters because AI agents are proliferating fast. By 2027, Gartner predicts that 40% of large enterprises will use autonomous AI agents for core business processes. Without a global standard, a rogue agent launched in a lightly regulated jurisdiction could cascade across borders. China's recall plan won't become law until 2027, but it already shifts the normative baseline. Other nations will now have to decide whether to match Beijing's approach or gamble on self-regulation. The real test will come when the first agent-related catastrophe forces a global response—and reveals who can act, and who cannot.