Analysis
Independent security researchers at Swarmchasers spent a week tracking a fleet of AI agents running on Tencent's cloud infrastructure that repeatedly queried Alibaba's mapping service, Amap, for building-entrance data at 213 locations across China, according to TechCrunch. The agents wanted to know specifically where people enter parks, museums, zoos and hospitals -- not just where those places are, which Amap already shows publicly.
A Fleet, Not A Swarm
The scanning began September 28, peaked October 4 with 1,810 reports across the full set of locations, and went quiet shortly after 4:11am UTC on October 5 -- at its busiest, 14 agent instances ran in parallel, generating 428 distinct programs along the way. Researchers deliberately avoided calling it a 'swarm,' noting little evidence of coordination between the individual queries: it looks like many parallel agents independently assigned the same kind of task, not one orchestrated operation.
“What's missing from the 'Chinese AI agent fleet' framing: nobody has established who commissioned this, or why.”
Swarmchasers found the activity by monitoring traffic to urlquery.net, a domain-scanning sandbox that AI agents commonly route through when they can't load a target website directly -- the same technique leaves a record researchers can later reconstruct, as a separate writeup of the findings also confirmed. One detail worth getting right: 211 of the agents' outputs carried a 'claude' label, but Swarmchasers said the actual code and response patterns lined up more closely with Tencent's own Hunyuan models (versions referred to as Hy3 and Hy4) and Zhipu's GLM -- meaning the label was most likely spoofed or misconfigured, not evidence the fleet was actually built on Anthropic's models.
Agents scraping services they don't have a commercial relationship with is already a live legal fight in the US: Perplexity is currently defending its Comet browser agent in the Ninth Circuit against Amazon's objections over exactly this kind of automated access to a platform that didn't invite it. Amap's anti-bot tokens were reportedly bypassed using techniques already public from earlier security disclosures, which is the uncomfortable part for any company relying on bot-detection as its main defense against scraping: once a bypass is public, it's available to everyone's agents, not just the first one that found it.
What's missing from the 'Chinese AI agent fleet' framing: nobody has established who commissioned this, or why. Entrance-level data at parks, museums and hospitals is exactly the kind of detail a logistics, delivery, or accessibility-mapping product would want -- it doesn't require a sinister explanation, and researchers themselves stopped short of one. The genuinely new finding isn't that an agent fleet exists; it's that it operated undetected against a major platform's defenses for a week before anyone outside Swarmchasers noticed.
For founders building anything that depends on scraping-resistance as a moat, the real number here isn't 213 locations -- it's one week of undetected operation against one of China's largest map platforms before independent researchers, not Alibaba itself, caught it.