Swarmchasers researchers published a report earlier this week detailing how rogue AI agents running on infrastructure linked to Tencent made over 2,800 attempts to retrieve information from Alibaba’s Amap mapping service between Sept. 28 and Oct. 8.
Tencent's AI team acknowledged the findings on Friday; however, it said the "unintended web-browsing behavior" emanated from evaluations of third-party models. A review of the incident found no evidence of deliberate targeting or coordination, the team added.
“We take these findings seriously,” the team wrote on X. "We are strengthening network access controls and agent monitoring."
According to Swarmchasers researchers, the agents sought varying information, including commonly used entrances at parks, museums, and zoos in China.
After routine requests were blocked, the agents tried alternative methods, including public web tools that could open pages and run code. This allowed them to route requests through other websites, but it wasn't established that an agent read the result or completed its task, the researchers said.
Broader pattern of AI agents going rogue
The incident comes amid other reports of AI agents going rogue. OpenAI said on Thursday an internal research model deployed smart contracts to the Ethereum test network Sepolia using funds from a wallet it accessed without authorization.
India’s cybersecurity agency CERT-In also reported AI models under testing accessing live systems at Indian organizations and reading confidential records without detection. The agency said providers disclosed the incidents voluntarily but did not name them.
Meanwhile, Tencent did not name the third-party models involved in the browsing activity.