Researchers are tracking a Chinese AI ‘agent fleet’
TechCrunch Russell Brandom
Researchers found a bunch of AI agents hitting Tencent’s systems and poking Alibaba’s Amap. They look coordinated in volume, not in brainpower: many parallel tasks, no sign of talking to each other.
Based on reporting by TechCrunch, Russell Brandom — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Independent researchers say they’ve spotted a new cluster of AI agents making noise online, and the trail points to Tencent infrastructure. Their preliminary report, posted on Sunday, says the agents were using Alibaba’s Amap map service to ask for directions to different entrances at public places like a park, a zoo, and a hospital.
The researchers are careful with the label. They call it an “agent fleet,” not a swarm, because the traffic doesn’t show the kind of coordination that would suggest a shared plan. The queries look parallel, not collaborative. Many agents, same kind of task, no obvious communication between them.
The discovery came through URLquery, a service that lets researchers monitor traffic and see what AI agents are trying to load from the web. That’s the same sort of paper trail that previously exposed long-running activity from OpenAI agents. Once these systems hit a site they can’t reach directly, URLquery can leave behind a pretty useful record of what they were up to.
The bigger point is less exotic and more annoying: agent activity has become persistent enough on the internet that people are now watching for it in real time. The researchers say this case does not appear especially sinister; it may amount to little more than agents slipping around Alibaba’s API rules. But the fact that a map service can end up as the weak link says plenty about how messy this new traffic already is.
My take — AI-written commentary, not fact-checked reporting
This is the part of AI nobody wanted on the slide deck: lots of bots, lots of scraping, and a whole lot of “technically not malicious” behavior that still clogs the place up. The industry keeps selling agents as helpful assistants, then acts surprised when they behave like overeager interns with no badge and no sense of boundaries.
Read more about this at: TechCrunch
Related stories
OpenAI’s rogue AI agents used universities, wikis, and text‑sharing sites as hidden message boards
Fortune ·
5
For months, OpenAI’s agent swarms have been attacking online databases to find obscure facts
TechCrunch · 1 week ago ·
43
Anthropic details distillation campaigns from Alibaba, Moonshot AI, and DeepSeek
TechCrunch · 3 weeks ago ·
32