LLMs are getting better at unmasking people online 

A recent study and reporting show that large language models can automatically deanonymize pseudo-anonymous online users by aggregating and correlating public, unstructured information (e.g., forum bios, social media), sometimes re-identifying individuals in minutes; the article cites research results, real-world examples (an AI doxxing incident), ethical limitations of the study, and warns that such capabilities lower the cost and increase the scale of privacy-invasive investigations by governments, companies, scammers, and cybercriminals.