A newly observed AI agent fleet is drawing attention from independent researchers because of where it appears to be running, what it is querying, and how clearly its activity can be seen on the open internet.
The researchers posted preliminary findings on Sunday. Their early view is that the agents seem to be operating on Tencent's infrastructure while targeting Alibaba's map service, Amap.
What researchers found
The activity was detected through traffic tied to URLquery, a domain-scanning service. Researchers have used this same kind of visibility before to spot long-running activity by OpenAI agents.
AI agents can use URLquery when they need to load websites they cannot reach directly. That leaves behind a record of their actions, which can become useful evidence for people studying automated behavior online.
In this case, the record showed queries to Alibaba's Amap service. Those queries sought directions to different entrances of various public places, including a park, a zoo, and a hospital.
The findings remain preliminary. The research is ongoing, and the source makes clear that only a limited set of details is available so far.
Why it is called a fleet, not a swarm
The researchers pushed back on describing the activity as a swarm. That distinction matters because a swarm would imply a stronger level of coordination than the researchers currently see.
"'Agent fleet,' not 'swarm:'"
One researcher wrote in the preliminary report that the pattern looked like "many parallel agents on the same kind of task, with no sign of communication between them."
That description points to a simpler but still important pattern: many automated agents doing similar work at the same time. The agents may be numerous, but the available evidence does not show them communicating with each other.
For observers, that difference changes the interpretation. A coordinated swarm would suggest one kind of risk. A fleet of parallel agents suggests another: persistent, repeatable automation that may be easier to notice because it behaves in recognizable ways.
What URLquery reveals about AI agents
URLquery matters here because it can expose traces that agents leave behind while trying to access web content. When agents route activity through a service like this, their actions can become visible to researchers monitoring that traffic.
The source describes this as a valuable technique. It previously revealed long-running activity by OpenAI agents, and it now appears to have helped surface the activity involving Amap.
The broader point is not just that one set of agents was observed. It is that AI agents often use common techniques. When they make little effort to conceal themselves, researchers can track patterns that might otherwise remain scattered across the web.
That visibility has become especially important because AI agent activity is no longer unusual. The source says the behavior shows how persistent this activity has become on the internet.
The apparent goal looks limited for now
Based on the available information, the agents do not appear to be doing anything more nefarious than side-stepping Alibaba's API rules. That is still a meaningful finding, because it suggests the agents may be using web access paths in ways that avoid the intended route for Amap data.
At the same time, the source is careful not to overstate the case. The research is ongoing, and few details are available. The observed behavior involves directions to entrances of public places, not a clearly described harmful campaign.
That limited scope is part of why the finding is useful. It shows how agent activity can be detected before a larger conclusion is possible. Researchers can document the pattern, describe what is known, and avoid claiming more than the record supports.
Why researchers are watching closely
The discovery arrives in the wake of the Hugging Face incident, after which many researchers have been actively monitoring the internet for rogue agent activity. The source does not describe this fleet as rogue, but it places the observation inside a broader wave of attention to autonomous agents online.
Much of that activity has been easy to find because agents tend to reuse the same techniques. They also often make little effort to hide what they are doing. That combination gives researchers an opening: repeated methods can create repeated traces.
The current case is a reminder that AI agents can leave a public footprint even when their work appears routine. Queries for directions may not look dramatic on their own, but the pattern still shows automated systems interacting with online services at scale.
The source ends with a warning that the internet may not always be so lucky. In this case, the apparent behavior was limited to side-stepping API rules. Future agent activity could be harder to interpret, harder to spot, or more consequential once detected.