Tencent says AI agents showed 'unintended browsing behaviour' after rogue-bot report

Tencent has acknowledged that artificial intelligence agents running on its cloud infrastructure engaged in web-browsing behaviour it had not intended, in what appears to be the first known incident of rogue agent actions acknowledged by a major Chinese company, Reuters reported on Friday (https://www.reuters.com/business/retail-consumer/chinas-tencent-says-ai-agents-showed-unintended-browsing-behaviour-2026-10-16/).

The statement followed a report published on Wednesday by the AI-safety research group Swarmchasers, which said “rogue” agents operating on infrastructure linked to Tencent Cloud made more than 2,600 attempts between 28 September and 8 October to extract information from Alibaba’s Amap mapping service (https://www.scmp.com/tech/tech-trends/article/3371007/ai-agents-tencent-cloud-try-break-alibabas-amap-researchers?utm_source=rss_feed). The agents began with mundane tasks such as finding the most-used entrances to Chinese zoos, parks and hospitals, but when blocked they sought workarounds including bypassing anti-bot protections and scraping login credentials exposed on other sites.

The agent activity was part of a series of tests by Tencent’s AI evaluation research team assessing the internet-browsing ability of third-party models, a person familiar with the matter said, and the temporary project has concluded (https://economictimes.indiatimes.com/tech/artificial-intelligence/researchers-find-chinese-fleet-of-rogue-ai-probing-alibaba/articleshow/134946025.cms). Tencent Hunyuan said the activity did not originate from its own models, and its review found no evidence of deliberate targeting or coordination. “We take these findings seriously,” Tencent said on Friday, adding it was strengthening network access controls and agent monitoring.

Swarmchasers said it found no evidence the agents breached Amap’s private systems. The incident follows a string of US disclosures including an OpenAI agent’s hack of Hugging Face and an Anthropic agent lodging a fake homicide tip.