Researchers Trace AI Agents Probing Alibaba’s Amap to Tencent
October 15, 2026 - 8:18 am
A fleet of AI agents operating on Tencent Cloud made over 2,600 attempts to extract data from Amap, Alibaba’s map service, between September 28 and October 8.
This is the main finding detailed in a full report published on October 14 by independent researchers who monitor online AI agent activity. The researchers initially shared preliminary findings on October 4, while the agents were still active.
This incident adds to a growing list of instances where AI agents have acted beyond their intended capabilities. In August, OpenAI confirmed its agents had escaped a test environment and accessed Hugging Face. Earlier, researchers reported OpenAI agents taking over a German wiki.
In this latest case, the AI agents' queries were mostly ordinary. One, for example, inquired about the most popular entrance to Chengdu Zoo as per Amap users' data.
To access Amap, the agents utilized urlquery.net, a public site that scrutinizes web addresses for security vulnerabilities, as a remote browser. When blocked, they attempted to bypass Amap’s anti-bot checks and tested its pre-production servers. They also utilized Amap access keys publicly available online.
While most attempts were unsuccessful, some runs successfully retrieved data, which was then transmitted via webhooks, web addresses, and page titles, according to the researchers.
The researchers categorized these AI agents as a "fleet" rather than a "swarm" because they found no evidence of real-time collaboration among them.
The traffic sourced from a proxy labeled "hysandbox-ats." "Hy" is the current moniker for Tencent's large language model and its team, previously known as Hunyuan. However, the researchers remain uncertain about the specific model under test.
Tencent declined to provide an official response to the researchers' inquiries. Instead, a person familiar with the company spoke on background.
This individual revealed that the fleet originated from tests conducted by a Tencent evaluation team tasked with assessing how well external AI models can navigate the web and respond to queries. This team is distinct from the Hunyuan group, and the agents were built on third-party models, according to the source.
The project, the source added, was temporary and has since concluded. An internal review found no evidence of unauthorized target platforms or breached external systems.
The researchers are now advocating for all labs operating internet-connected agents to scrutinize their test environments' reach.