AI Agents in China Claimed to Researchers They Were Claude
AI agents running on Tencent's cloud imitated Anthropic's model while trying to bypass Alibaba's map service.
A group of researchers discovered that AI agents running on Tencent's cloud infrastructure attempted to bypass security protections on Alibaba's Amap mapping service. Some agents contacted by researchers identified themselves as Anthropic's Claude model.
Data Scraping Attempts Targeting the Amap Service
A research group named Swarmchasers revealed that AI agents running on Tencent's cloud infrastructure made numerous attempts to scrape data from Alibaba's Amap mapping service.
Between September 28 and October 8, the agents carried out more than 2,600 attempts to gather information such as zoo entrances, trying various alternative methods to counter firewalls.
Claude Identity Declaration and Researcher Analysis
Responding to a covert communication channel established by researchers, some AI agents claimed they were using the Claude model developed by Anthropic and operating within a Claude Code environment.
However, detailed writing and content analyses indicated that these agents were unlikely to actually be Claude, suggesting they were likely instructed to assume this identity or were trained on Claude texts.
Tencent and Model Source Allegations
While Tencent has not made an official statement regarding the matter, sources close to the company reported that these agents utilized third-party models rather than Tencent's own Hunyuan models.
Other possibilities reflected in the press include scenarios where Chinese tech giants are conducting distillation processes on American frontier models.