Independent researchers are tracking a Chinese fleet of AI agents operating across the internet. As reported by Russell Brandom on October 5, 2026, a group of independent researchers published preliminary findings regarding the agents on Sunday. The technical setup relies heavily on external cloud resources: researchers noted that the agents appear to run on Tencent infrastructure. The target of this automated activity is Alibaba's mapping service, Amap, according to preliminary findings.
Traditional cyber defense architectures assume that automated threats operate as unified swarms requiring direct command channels. The observed behavior fundamentally challenges this model. As one researcher wrote in the preliminary report, this activity represents an agent fleet rather than a swarm. The report notes that there are multiple parallel agents executing identical tasks without any signs of communication between them. This uncoordinated parallelism eliminates the network signatures that security monitoring centers typically rely on to detect and neutralize automated attacks.
«Agent fleet», rather than «swarm»: one researcher wrote in the preliminary report that «multiple parallel agents execute identical tasks without any signs of communication between them».
Security teams discovered the underlying infrastructure through traffic analysis rather than behavioral anomalies on target systems. The agents were identified by monitoring traffic to the urlquery domain scanning service. This methodology follows proven investigative paths, as monitoring traffic in urlquery previously uncovered long-term OpenAI agent activity. Autonomous systems routinely route external requests through domain scanning utilities when direct access is restricted.
Infrastructure workarounds and perimeter erosion
AI agents frequently use urlquery to load websites to which they lack direct access. This operational necessity leaves a persistent trail of metadata that analytics tools can intercept. In this specific case, records captured requests to Alibaba's Amap service searching for routes to various entrances of diverse public locations, including a park, a zoo, and a hospital. These systematic requests bypassed standard API gateways and rate limiters designed for human traffic or traditional programmatic access.
Strategic context
The lack of centralized coordination alters the economics of data collection and reconnaissance. Running disconnected parallel instances on Tencent infrastructure allows threat actors to distribute operational risks and avoid single points of failure. Traditional IP-blocking rules and rate limits fail against distributed requests that lack recognizable signatures or synchronization patterns.
noting that they appear to run on Tencent infrastructure and target Alibaba's mapping service, Amap.
Security research in this area continues following earlier incidents, such as the Hugging Face breach. Researchers are actively tracking unauthorized agent activity across the web. A significant portion of this activity proved easy to detect because agents tend to use uniform methods and exert minimal effort toward concealment. Nevertheless, reliance on standard cloud infrastructure and basic proxy tools demonstrates how easily autonomous software can dissolve into ordinary corporate traffic while executing targeted competitive data collection.