LLM-Assisted Deanonymization

A collection of security posts highlighting AI-driven and traditional threats. Key items: research shows LLMs can deanonymize users from a handful of anonymous posts and then re-identify them at web scale; LLMs also produce highly predictable (and therefore weak) passwords; poisoning of AI training data is trivially easy via web content and can rapidly corrupt model outputs; researchers found practical server-side/backdoor pathways in popular password managers that can expose vaults or weaken encryption; North Korean operators are running fake recruiter/coding-challenge campaigns that install‌