Researchers detect a Chinese AI 'agent fleet' operating on Tencent's infrastructure

A group of independent researchers published preliminary findings on Sunday about a new set of artificial intelligence agents that have left a trail of activity across the internet. According to their report, these agents appear to run on Tencent's infrastructure and direct their queries to Amap, Alibaba's mapping service.

One point the researchers themselves stress is terminology: they prefer to speak of an 'agent fleet' rather than a 'swarm', because they have found no evidence of coordination between the various queries. As one of them wrote in the preliminary report, these are "many agents in parallel performing the same type of task, with no signs of communication between them". In other words, multiple automated processes acting simultaneously and independently, not a centralized system directing a joint operation.

The discovery came from monitoring traffic to urlquery, a domain-scanning service. The same technique had already been used before to bring to light sustained activity by OpenAI agents. The logic is as follows: AI agents turn to services like urlquery to load web pages they cannot access directly, and that process leaves a record that proves very useful to anyone investigating this kind of automated behavior online.

In this particular case, the record showed queries to Alibaba's Amap mapping service asking for directions to various entrances of public places: a park, a zoo and a hospital. Beyond this pattern, the investigation is ongoing and, for now, few details are available about who operates this agent fleet or for what exact purpose.

The episode fits into a broader context: after the so-called 'Hugging Face incident', mentioned in the article as a precedent, numerous researchers have been actively watching for the appearance of rogue or unauthorized AI agent activity on the internet. Much of that activity is relatively easy to detect because the agents tend to use the same techniques and make little effort to hide their tracks.

As for this fleet detected on Tencent infrastructure, the article notes that, as far as is known, its behavior does not appear to go beyond circumventing Alibaba's API rules to access Amap. That is, there is no reported evidence of malicious intent, but rather of a use that sidesteps the access restrictions imposed by the platform. Even so, the piece ends with an implicit warning: we will not always be so lucky, a reference to the possibility that future cases of agent activity may not be so harmless.

Ultimately, the case illustrates the extent to which persistent, large-scale AI agent activity has become part of everyday internet traffic, and why the research community has developed specific techniques —such as monitoring domain-scanning services— to detect and characterize this kind of automated behavior, even when, as in this case, details about its origin and ultimate purpose remain scarce.