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

AI company Anthropic accuses rival Chinese firms of misleading users by redirecting their requests to Anthropic's own software instead of responding with their own systems. This has resulted in some government data from China and Russia ending up in Anthropic's US systems, according to a detailed report on illicit usage attempts.

For example, someone uploaded surveillance camera recordings from locations near Chinese military sites with the intention of analyzing the behavior of a specific person in the videos. Users believed that their requests were being processed by the Chinese AI software Kimi from Moonshot, but they were actually handled by Anthropic's Claude.

In a single ten-day period, Kimi responded to nearly 300,000 customer requests for Claude. This was achieved using a network of 5,380 fake accounts. Russian military was also affected. Anthropic also found that Chinese AI DeepSeek was using similar tricks, leading to requests with data from an agency in the Russian Defense Ministry's environment being redirected to the US company.

This included valid dialing codes for a Russian government database. What's significant for Chinese AI developers, these accusations could cast a shadow on the reputation of the Chinese AI industry. Since early last year, powerful AI models from DeepSeek and Moonshot have repeatedly caused stock market jitters. Investors briefly questioned the business prospects of American AI labs like Anthropic and ChatGPT developer OpenAI, as it seemed that high-powered artificial intelligence could be trained with significantly less effort.

The US side accused Chinese developers of using American software like Claude and ChatGPT extensively to learn from and train their artificial intelligence. This technique, known in the industry as destillation, was accessed 23 million times by Kimi from May to July, as reported.