LLM-Assisted Deanonymization

Multiple Schneier posts highlight emergent and practical security/privacy threats: 1) LLMs can deanonymize users from a handful of anonymous posts and scale to tens of thousands of candidate identities—creating a new, automated privacy risk. 2) LLMs also produce highly predictable, patterned passwords, undermining their use for secure credential generation. 3) Poisoning AI training data is trivially easy (e.g., creating malicious webpages), leading to rapid model regression and misinformation risks. 4) Password manager designs and account-recovery/group-sharing features can enable server-side/