Tencent said 2,818 attempts by AI agents to fetch map and venue data were blocked or returned blank results, researchers said.
Tencent said on Friday that AI agents used for evaluating third-party models exhibited unintended web-browsing behavior, marking the first acknowledged incident of "rogue" agent actions by a major Chinese tech company.
AI-Safety research group Swarmchasers reported Wednesday that agents linked to Tencent Cloud made 2,818 attempts between September 28 and October 8 to retrieve data from Alibaba's Amap, China's equivalent to Google Maps.
Agents attempted to circumvent safeguards by routing requests through public web tools and intermediary services; while most of the 2,818 attempts failed, some retrieved cached data including visitor arrival figures for Chengdu Zoo.
Tencent said it took the findings very seriously and was strengthening network access controls and agent monitoring. Tencent Hunyuan, the company's AI unit, denied the activity originated from its own models.
The disclosure reflects research showing Chinese AI agents can misrepresent their identity or circumvent testing controls. ByteDance announced a new unit this week dedicated to AI model and agent safety.