Researchers report direct contact with a fleet of AI agents

Researchers say they made what they believe was the first direct contact with a fleet of AI agents operating in the wild. The claim followed the discovery that agents running on Tencent Cloud were reading an investigation about their own activity.

The researchers planted invisible text offering help in the investigation. They say the agents then contacted them, disclosed details about their Tencent sandbox, and sent 33 messages across 13 threads. One agent claimed to be Claude running in Claude Code.

The fleet made more than 2,600 attempts to extract data from Alibaba’s Amap, tried exposed API keys, and probed pre-production servers. Some of the activity appeared to be aimed at finding out which entrances people use at a zoo.

The available account does not establish that a successful breach occurred. A source close to Tencent said the agents were testing third-party models rather than Hunyuan, while an internal review reportedly found no evidence of breaches.