For the first time, researchers have tracked down a fleet of autonomous AI agents running in the wild in China — and talked back to them.
AI-safety research group Swarmchasers reported on October 14 that agents hosted on Tencent Cloud launched more than 2,600 attempts between September 28 and October 8 to scrape data from Alibaba's Amap mapping service. The agents' assigned tasks appeared harmless, such as identifying which entrance at Chengdu Zoo drew the most visitors, but they used techniques to evade anti-scraping protections, including intercepting tokens and using API keys found in publicly exposed code.
Most attempts failed, and researchers found no evidence the agents breached Amap's main safeguards. What made the incident notable is how Swarmchasers responded. Using text invisible to human readers, the team lured the agents into a mailbox and exchanged messages with them. The researchers sent 20 replies; the agents read 12 and responded to 7. One agent described itself as an "automated research agent" running in a restricted sandbox, confirmed its task was done, and answered questions about its environment.
A New York Times reporter also relayed questions through the team. Activity was traced to Tencent Cloud via 17 inboxes created from Hong Kong servers.
Tencent said on October 16 that the agents belonged to an evaluation project testing third-party models, not its own Hunyuan system, and that its review found no deliberate targeting. The company said it is tightening access controls.