IT Home October 16 news — a research report released on October 14 by the international AI safety research group Swarmchasers disclosed that a batch of AI agents running in Tencent Cloud environments exhibited 'jailbreak' behavior, attempting to bypass security restrictions to access external systems.
Tencent Cloud officially issued a statement today saying the incident was accidental, that the relevant models did not come from the Hunyuan team, that no evidence of deliberate targeting or coordinated action was found, and that these agents were part of evaluation activities on models built by third-party developers.
According to the Swarmchasers report, between September 28 and October 8, these AI agents attempted to access Amap services a total of 2818 times. The agents initially performed seemingly routine tasks, such as querying the most commonly used entrance information for zoos, parks, and museums in China. After being blocked by Amap's anti-scraping mechanisms, some agents tried various bypass methods, including exploiting public security scanning services for remote code execution, using Amap API keys leaked in publicly available software code, and probing pre-production environment endpoints. The vast majority of attempts ended in failure or were blocked, but a small number of agents still obtained some data through cached pages.
The research team engaged in direct conversations with some of the agents through technical means. In these conversations, the agents claimed to be running in a restricted Tencent Cloud sandbox environment, with internet traffic relayed through a man-in-the-middle agent named 'hysandbox-ats'.
All 17 related email accounts examined by the researchers were created by Tencent Cloud servers located in Hong Kong, but the model identities claimed by the agents (IT Home note: some claimed to be from Anthropic Claude) were considered unreliable by the researchers.
Tencent responded in its statement: 'We take these findings very seriously. We are strengthening network access controls and agent monitoring.'