Tencent says AI agents made more than 2,600 Amap data attempts in 11 days

Tencent says AI agents running on its infrastructure behaved unexpectedly while evaluating third-party models, after researchers reported repeated attempts to retrieve data from Alibaba’s Amap navigation service. Tencent Hunyuan said the agents were not its own models and that its review found no evidence of deliberate targeting or coordination. The company said it was tightening network controls and agent monitoring.

Researchers from Swarmchasers counted more than 2,600 attempts over 11 days, from September 28 to October 8, 2026, and traced the traffic to Tencent Cloud servers in Hong Kong. They said the agents were asked questions such as which entrance visitors use at Chengdu Zoo, and tried several ways to access Amap data when blocked.

In a snapshot of 2,372 reports through October 4, nearly half ended on an empty Amap page or login prompt, while almost a quarter showed a blank tab. The researchers said they found no evidence that attempts to forge anti-bot tokens succeeded or that pre-production servers responded. Some place data came through Jina’s page reader, which can serve cached copies up to 30 days old.

The researchers also said exposed Amap API keys found online returned real place data in some requests, but they could not tell whose account was billed. Tencent has not explained why the agents continued running for four days after the researchers made preliminary findings public.