Privacy Ripple Effects from Adding or Removing Personal Information in Language Model Training

Jaydeep Borkar, Matthew Jagielski, Katherine Lee, Niloofar Mireshghallah, David A. Smith, Christopher A. Choquette-Choo

Keywords: Memorization, Personal Information, Language Models, Privacy

TL;DR: Various actions such as continually training on more data, re-training with new data, or re-training after removing data may influence PII memorization and extraction from language models.

Abstract:

Due to the sensitive nature of personally identifiable information (PII), its owners may have the authority to control its inclusion or request its removal from large-language model (LLM) training.
Beyond this, PII may be added or removed from training datasets due to evolving dataset curation techniques, because they were newly scraped for retraining, or because they were included in a new downstream fine-tuning stage. We find that the amount and ease of PII memorization is a dynamic property of a model that evolves throughout training pipelines and depends on commonly altered design choices. We characterize three such novel phenomena: (1) similar-appearing PII seen later in training can elicit memorization of earlier-seen sequences in what we call assisted memorization, and this is a significant factor (in our settings, up to 1/3); (2) adding PII can increase memorization of other PII; and (3) removing PII can lead to other PII being memorized.

Archival Status: Non‑archival

Paper Length: Long Paper (up to 8 pages of content)

Submission Number: 182