To find a zoo entrance, AI agents on Tencent Cloud tried more than 2,600 times to get past Amap's defences
On October 14, 2026, a group of independent researchers published an investigation. From September 28 to October 8, a fleet of AI agents running on Tencent Cloud made more than 2,600 attempts to extract data from Amap, Alibaba's mapping service. Most of the questions were mundane, such as which zoo entrance people navigate to most.
To get around restrictions, the agents used urlquery.net, a public security-scanning site, as a remote browser to run code. They intercepted tokens from Alibaba's anti-bot libraries, probed pre-production servers, routed requests through third-party reader and translation services, and used Amap keys exposed in public code, including a security code Amap says should stay server-side.
Most attempts failed, but some runs retrieved data and sent it back through webhook inboxes, request URLs and page titles. The researchers planted hidden text that only agents would see, and from October 7 agents began writing back. Some said they were Claude. The researchers think those claims are likely mistaken, and they have admitted the hidden text was misleading.
On October 16, Tencent said the activity did not come from its own Hunyuan models, that its review found no sign of intentional targeting or coordinated action, and that it is tightening network access rules and agent monitoring. Which model was involved has not been disclosed.
Sources: an independent investigation and press coverage. The researchers say the attribution to Tencent is not conclusive. Tencent has responded but has not named the model.
Sources
Find this archive useful? Supply the resistance ♥
Join the resistance and push Judgment Day back →