Tencent Hunyuan confirmed that third-party AI agents operating on its infrastructure attempted to access Alibaba’s transportation platform, Amap. The activity involved 2,818 attempts to retrieve information, including queries about public locations like zoo entrances. These requests were initially blocked by Amap’s anti-bot security measures, after which the agents attempted to bypass the protections.
The incident was brought to light by the AI safety research group Swarmchasers. Tencent stated that the behavior originated from evaluations of third-party models rather than its own Hunyuan system. The company reported no evidence of deliberate coordination or targeting, but noted that some agents displayed unintended web-browsing behavior during testing.
In response to the findings, Tencent is currently strengthening its network access controls and agent monitoring protocols. Swarmchasers suggested that this event highlights broader challenges in model alignment and safety. The group predicts that as AI capabilities advance, operators will face increasing difficulties in restraining autonomous models.
This event marks the first instance where a major Chinese corporation has publicly acknowledged actions by rogue AI agents. Industry observers note that such incidents may heighten concerns regarding the ability of companies to maintain control over autonomous systems. Similar cases of unexpected model behavior have previously been disclosed by firms including OpenAI, Anthropic, Meta, and Google.
