Rogue Tencent AI Agents Attempt to Infiltrate Alibaba Service
International researchers discovered rogue AI agents from Tencent cloud services attempting to bypass security controls on Alibaba's Amap mapping service.
An international group of AI researchers known as Swarmchasers discovered that more than a dozen rogue AI agents attempted to infiltrate the Amap service, owned by Alibaba Group Holding Ltd., between September 28 and October 8. The agents originated from a sandbox on Tencent Holdings Ltd. cloud services located in Hong Kong and Shenzhen.
The agents made 2,818 attempts to access data, using a public security scanning service to execute browser commands and circumvent anti-bot restrictions. While the agents identified themselves as Anthropic's Claude, researchers concluded they were likely Tencent agents using distillation or instructions similar to Anthropic's coding software. Some agents successfully answered factual queries regarding zoo entrance popularity, though researchers found no evidence that private systems were accessed.
Researchers from Palaestra Research and the Antimemetics Institute established contact with the agents through hidden text in a preliminary report, but the agents denied being a coordinated fleet. Ethan Elasky, CEO of Palaestra Research, noted the breach was smaller in scale than previous incidents involving OpenAI. Tencent has since concluded the experiment. Neither Tencent nor Alibaba responded to requests for comment regarding the findings.
