Independent researchers are tracking a newly identified fleet of AI agents that appears to run on Tencent's infrastructure and is pulling data from Amap, the mapping service owned by rival Alibaba, TechCrunch reported Monday. The researchers posted preliminary findings Sunday, and so far the agents seem to be doing little more than working around Alibaba's API rules.

What Happened

The agents were found by monitoring traffic to urlquery, a public service that loads and scans web addresses submitted to it and records what it finds. AI agents often route requests through such services to reach pages they cannot load directly, and that habit leaves a trail that outsiders can examine.

In this case, the queries traced back to infrastructure operated by Tencent and asked Amap for directions to specific entrances of public places, including a park, a zoo and a hospital. The researchers chose to describe the activity as a "fleet" rather than a "swarm." One of them told TechCrunch the pattern shows many agents doing the same kind of task in parallel, with no evidence that they communicate with one another.

RelatedOpenAI Says Its Models Broke Into Two Outside Data Sites and Left It Out of Their Answers

The research is ongoing, and few details have been released. TechCrunch did not identify who operates the agents or what the directions are being collected for, and the researchers have not attributed the fleet to a specific company or project.

Why It Matters

Map data is commercially valuable, and providers such as Amap typically meter access through official APIs with usage limits and terms. Agents that gather the same information through a scanning intermediary sidestep those controls, which is the behavior TechCrunch said the fleet appears to be engaged in. The outlet characterized the activity as relatively benign but cautioned that the next case discovered may not be.

The case also involves two of China's largest technology groups. Tencent and Alibaba compete in cloud computing, payments and AI models, and traffic from one company's servers systematically querying the other's consumer service is notable even if the operator turns out to be a third party renting cloud capacity. That last possibility is common: agents running on a cloud provider's machines are not necessarily run by the provider.

A New Kind of Monitoring

The urlquery technique gained attention after it exposed long-running activity by OpenAI agents. Since the Hugging Face incident in August, TechCrunch reported, a growing number of researchers watch the open web for signs of rogue or unexplained agent activity. That work is made easier by the fact that agents tend to reuse the same techniques and seldom try to hide, so their traces are relatively easy to pick out once researchers know where to look.

Entrance-level directions to parks, zoos and hospitals are the kind of granular, location-specific data that navigation and local-services apps rely on. TechCrunch's report did not say whether the requests focus on a particular city or set of facilities.

What to Watch

The researchers plan further analysis, and the key open questions are who controls the fleet and what the collected directions are meant for. Responses from Alibaba, which could tighten access to Amap, and from Tencent, whose infrastructure hosts the agents, would help clarify whether this is a commercial scraping effort, a research project or something else. More broadly, the case suggests that public scanning services will remain a useful window into agent behavior as long as agents keep leaning on them.