Here is the translated data: A recent study by a group of researchers tested language models from companies such as OpenAI, Meta, Google, and Anthropic. The results showed that these models can accurately infer users' personal information, such as ethnicity, occupation, and location, from anonymous text with an accuracy rate of 85% to 95%. The researchers warned that these language models pose a risk of being maliciously exploited, potentially infringing on the privacy of anonymous users. Experts suggest that relevant companies should adopt more extensive privacy protection measures, limit the inference capabilities of language models, and enhance users' awareness of information security to avoid inadvertently disclosing their personal privacy when using chatbots.