Independent researchers have published preliminary observations of multiple AI agents apparently making route queries to Alibaba’s Amap service. TechCrunch reports that the agents seem to operate on Tencent infrastructure, while stressing that few details are available and the investigation is continuing. The visible activity concerns directions to entrances at public places, not a demonstrated coordinated attack. The researchers prefer the description agent fleet because they saw parallel work on a similar task without evidence that the agents were communicating with one another.
Similar queries do not establish a coordinated swarm
The preliminary findings were posted on Sunday, according to TechCrunch’s October 5 account. Researchers observed activity directed at Amap, Alibaba’s map service, with queries seeking routes to different entrances of places including a park, a zoo and a hospital. Those examples describe the tasks visible in the traffic record. The source does not explain the broader purpose behind the requests or identify a completed service built from them.
The apparent connection to Tencent is an infrastructure observation. TechCrunch says the agents seem to be running there, rather than presenting a confirmed account of who commissioned or operated them. Treating that association as proof that a particular company directed the activity would go beyond the attribution available in the report. The label Chinese agent fleet should be read alongside that limitation.
The researchers’ objection to swarm is more than a choice of headline language. Their observation was of many agents working in parallel on the same kind of task, with no sign of communication between them. Similar destinations or queries do not by themselves show a shared control mechanism. The available account therefore supports parallel activity, while leaving coordination unestablished.
An intermediary service provides a window into agent behaviour
The researchers found the activity by watching traffic to URLquery, a domain-scanning service. TechCrunch explains that AI agents often use it to load websites they cannot access directly. That process leaves a record of requests, giving observers a way to examine some agent activity that might otherwise be difficult to see.
The article notes that the same monitoring technique previously exposed long-running OpenAI agent activity. That earlier example is context for the method, not evidence that the newly observed agents belong to OpenAI or have the same objectives. In this case, the reported record points to Amap queries and an apparent Tencent infrastructure connection. Those are the specific observations at the centre of the new account.
A request record is also narrower than a full explanation of an agent system. The report provides examples of destinations and describes the monitoring route, but does not supply a detailed account of the underlying software or instructions. Readers can separate what researchers saw being requested from questions about why the requests were made, how the agents were launched and what they ultimately produced.
The findings remain preliminary rather than evidence of an attack
TechCrunch describes the activity as apparently doing little more than sidestepping Alibaba’s API rules. That assessment does not establish a malicious campaign, and the source reports no demonstrated destructive outcome from these queries. Amap requests to public-place entrances should not be reframed as evidence of a cyberattack simply because AI agents made them.
The wider monitoring effort follows concern about rogue agent activity after the Hugging Face incident. According to TechCrunch, researchers have found some activity relatively easy to detect because agents reuse techniques and make little effort to hide. That is background to the current investigation, not a finding that every visible agent is rogue or that this fleet caused the earlier incident.
The present contribution is a limited but specific observation: parallel agents, an apparent infrastructure association and a series of map-service queries visible through an intermediary. The continuing investigation may clarify those relationships. Until it does, the distinctions between observed traffic, inferred operation and unproven intent are essential to understanding what the preliminary report actually shows.
