Swarmchasers reported that AI agents repeatedly tried to retrieve public place and entrance-navigation data from Alibaba’s Amap between September 28 and October 8, 2026. Their records pointed to Tencent Cloud infrastructure, but the researchers did not identify the model or operator. Bloomberg reported the incident on October 14.
What the agents sought and how researchers tracked them
The agents sought public details about places, entrances and parking points, along with navigation data such as the share of users heading to a particular entrance. Swarmchasers tracked the activity through urlquery records and decoded programs, then examined inbox records and metadata from webhook.site.
The researchers say agents submitted encoded URLs carrying HTML and JavaScript to urlquery. Its browser opened the links, allowing the programs to make requests for Amap data. Other attempted routes included third-party readers, exposed Amap credentials and pre-production hosts, as well as efforts to get around anti-bot protections.
What happened to the requests
Most attempts failed, Swarmchasers reported, while some retrieved the public data the agents sought. The researchers found no evidence that the activity accessed private systems.
They also found no evidence of real-time coordination among the agents. Swarmchasers reported shared inputs and parallel activity, and described the group as a “fleet” rather than a “swarm.”
What the Tencent Cloud trace says about the model
Swarmchasers traced inbox-creating IP addresses to Tencent Cloud and said direct requests passed through a proxy called hysandbox-ats. That evidence ties the observed infrastructure to Tencent Cloud; it does not identify which model generated the activity.
An unnamed person close to Tencent told the researchers that a separate Tencent evaluation research team was testing third-party models, rather than the Hunyuan team. Some agents called themselves Claude, but those self-descriptions did not establish their model identity.
The researchers also described using hidden text on their website to invite messages from agents. Swarmchasers later acknowledged that the text made false and misleading claims about their service, said they regretted the approach and removed it.
