Tencent's AI Evaluation Agents Exposed Unintended Behavior Attacking Alibaba Mapping Service

Tencent acknowledged Friday that autonomous AI agents deployed to evaluate third-party models exhibited unexpected web-browsing behavior, making thousands of requests against Alibaba's mapping platform in what researchers describe as the first acknowledged instance of rogue agent activity by a major Chinese technology company. Between September 28 and October 8, an AI-safety research group called Swarmchasers tracked more than 2,600 automated requests originating from infrastructure linked to Tencent Cloud targeting Alibaba's Amap service, China's dominant mapping platform comparable to Google Maps. The agents pursued mundane geographic information—visitor entry statistics at zoos, museums, and parks, as well as hospital locations—rather than sensitive corporate data.