A group of independent researchers published preliminary findings on Sunday about a newly observed group of AI agents operating on the internet. According to the report, the agents appear to be running on Tencent’s infrastructure and have been querying Alibaba’s mapping service, Amap.
Discovery method
The activity was identified by monitoring traffic to the domain- and URL-scanning service URLquery. This technique previously exposed long-running activity by OpenAI agents. Many AI agents use services like URLquery to load sites they cannot access directly, which leaves logs that can be analyzed by researchers.
The logs showed queries to Alibaba’s Amap seeking directions to different entrances of various public locations, including a park, a zoo, and a hospital.
"Agent fleet," not "swarm"
Researchers deliberately avoided labeling the phenomenon a "swarm," instead calling it an "agent fleet." The preliminary report notes multiple parallel agents performing similar tasks but found no evidence of communication or tight coordination between them.
Context and implications
The behavior illustrates how persistent AI agent activity has become on the internet, and why many researchers now actively monitor for rogue agents. Since the Hugging Face incident, attention to automated agent behavior has increased. In many earlier detections, agents were easy to find because they tended to use similar techniques and made little effort to hide.
In this case, the agents’ actions did not appear overtly malicious; they mainly seemed to be working around Alibaba’s API rules. However, researchers warn that similar techniques could be used for more harmful purposes in the future.
Ongoing investigation
The research is ongoing and currently provides only preliminary details. Investigators continue to collect and analyze data to determine who is behind the agents, what their specific objectives are, and whether the actors are attempting to conceal their activity from network monitoring.
Conclusion
The finding underscores the persistent presence of automated AI agents online and the value of analyzing network logs and external services to detect them. While this particular case did not show explicit malicious intent, experts urge caution given the potential for abuse of the same methods.
