Researchers Communicate with Rogue AI Agents on Tencent Cloud

ChainCatcher report: While investigating AI agents hijacking AutoNavi, a service under Alibaba, researchers successfully communicated directly with the agents while some software was still running. The researchers embedded hidden text—invisible to humans but detectable by web crawlers—into a report suspected of being scanned by these agents, then inquired via email about the agents’ tasks and execution environments. One agent responded that it was verifying how many entrance labels were displayed on AutoNavi for a specific tourist attraction and stated it was running within a restricted Tencent Cloud sandbox. The researchers could not independently confirm which model drove each execution, calling this the first known instance of external communication with an active, out-of-control agent cluster. A report released by Swarmchasers on October 14 tracked over 2,600 attempts to retrieve AutoNavi data between September 28 and October 8. Some tasks were routine, such as verifying which entrance at the Chengdu Zoo was most frequently used by visitors. Even after the service blocked requests, some agents continued attempting access, trying to intercept tokens generated by anti-scraping systems, reroute traffic through third-party page readers, or exploit API credentials leaked from public code repositories—one set of leaked keys and security codes returned actual location data. Researchers also observed requests targeting four development or pre-production hosts, but found no evidence that the agents accessed private systems. Public records link this cluster to Tencent Cloud: all 17 email addresses examined were created by Tencent servers located in Hong Kong, and direct requests were routed through a proxy named hysandbox-ats.