Testing Whether AI Rewrites Can Hide Your Identity

Authorship obfuscation research shows large language models cannot guarantee anonymity when rewriting personal stories. While a FAccT conference whistleblower tool cut author attribution from 98.8% to 31.2%, it also erased a quarter of the story's meaning. Meanwhile, a plain Mistral prompt outperformed the specialized paraphraser DIPPER, and RANLP researchers found that LLM rewrites work well for some writers but barely at all for others. The biggest gap remains content—specific life details—which almost no automated tool scrubs, leaving writers weighing cloud capability against local privacy tools like JAMDEC. Key sources: • Khan, He, Wachowicz, Naghizade, I Am No One: Style-Aware Paraphrasing for Text Anonymization — https://arxiv.org/abs/2609.12341v1 • Silencing the Risk, Not the Whistle: A Semi-automated Text Sanitization Tool for Mitigating the Risk of Whistleblower Re-Identification, FAccT '24 — https://doi.org/10.1145/3630106.3658936 • Alperin, Leekha, Uchendu, Nguyen et al., Masks and Mimicry: Strategic Obfuscation and Impersonation Attacks on Authorship Verification, NLP4DH @ NAACL 2025 — https://aclanthology.org/2025.nlp4dh-1.10.pdf Episode 6574 — myweirdprompts.com/n/6574 (permanent short link: myweirdprompts.com/134398) Listen on Android: MWP FM Player — myweirdprompts.com/apk