AI agents are now a persistent background noise on the internet, and researchers are starting to map the signal. According to TechCrunch, a group of independent researchers posted preliminary findings this past Sunday about a fleet of AI agents operating on Tencent’s infrastructure and directing queries at Alibaba’s mapping service, Amap. The behavior appears benign so far. But that’s almost beside the point.
The agents were caught through traffic monitoring on urlquery, a domain-scanning service that has become a quiet but reliable window into agent activity. Agents frequently use urlquery to load websites they can’t access directly, and that habit leaves a paper trail. The same method previously exposed long-running activity by OpenAI agents. In this case, the logs showed repeated queries to Amap, asking for directions to different entrances of public places like parks, zoos, and hospitals. Parallel queries, no apparent coordination between them, no obvious goal beyond the task itself.
The researchers were deliberate about terminology. They specifically rejected the word ‘swarm’ because swarms imply coordination. What they observed was something different: many agents running the same kind of task independently, with no sign of communication between instances. They called it an ‘agent fleet.’ That distinction matters because it shapes how we think about risk. A swarm is directed. A fleet might just be poorly governed.
This fits into a broader pattern that has been building since the Hugging Face incident earlier this year, which put the research community on alert for rogue agent activity. What researchers have found since then is that most agents are easy to spot because they reuse the same techniques and make almost no effort to hide. That’s reassuring in one sense. It also means the ones that do try to stay hidden are much harder to find.
In competitive context, this is a Chinese infrastructure story as much as it is an agent story. Tencent and Alibaba are both deep into AI buildout, and agent-based products are a major focus across the Chinese AI market, alongside companies like Baidu and ByteDance. The specific combination here, agents on Tencent’s cloud hitting Alibaba’s mapping API, suggests someone was either testing a product, sidestepping API access rules, or running a data collection job without bothering to clear it through official channels.
The research is still ongoing, so the full picture isn’t clear yet. But the practical takeaway is already visible: AI agents are operating continuously across public infrastructure, often without oversight, and the main thing catching them right now is their own sloppiness. That won’t hold indefinitely.
