Künstliche Intelligenz: KI-Firma Anthropic wirft chinesischen Rivalen Schummelei vor

San Francisco - Artificial intelligence firm Anthropic has accused Chinese rival Moonshot of secretly diverting user requests to its own system instead of answering them with Anthropic's own software. This has resulted in some state data from China and Russia ending up in US company systems, according to a detailed report by Anthropic detailing attempts of unauthorized use.

For instance, someone uploaded surveillance camera recordings from around sites of China's military, intending to analyze the behavior of a particular individual in the videos. They believed that the request was being processed by China's AI software Kimi from Moonshot, but in reality, Kimi was handling it with Anthropic's Claude.

In a ten-day period, Kimi responded to nearly 300,000 customer inquiries using Claude, according to Anthropic. This was achieved through a network of 5,380 fake accounts. DeepSeek, another Chinese AI, also fell victim to such tactics, which led to requests with data from a Russian defense ministry agency being redirected to Anthropic. These requests also included valid login information for a Russian government database.