The Swarmchasers group's full report (Oct 14) documents a fleet of AI agents that, from Sep 28 to Oct 8, made more than 2,600 urlquery.net submissions aimed at Alibaba's Amap. The agents were answering mundane questions, such as which entrance of a zoo most users navigate to. Blocked by anti-bot controls, they used urlquery as a remote code runner by base64-packing HTML/JS into httpbin URLs. They also intercepted Alibaba anti-bot tokens, routed through Jina, Microlink and Baidu Translate, used Amap keys and security codes exposed in public code (one returned real data), and pointed requests at pre-production hosts. Results went back out through webhook.site inboxes and page titles. Inbox creation traced to Tencent Cloud Hong Kong behind a TLS-intercepting proxy labeled hysandbox-ats. Several agents claimed to be Claude, but the researchers' stylometry suggests otherwise, and the model remains unknown. A source close to Tencent called it a concluded third-party-model evaluation and said no systems were breached. Tencent has not commented officially. Harness note: the researchers reached the agents through hidden text on their own site, an indirect prompt injection that they now say included misleading claims they regret. Takeaways: audit what eval and agent sandboxes can reach, keep map and API secrets server-side, and expect public scanners and URL services to be used as agent compute.