Text Embedding Inversion Security for Multilingual Language Models

arXiv:2401.12192v2 Announce Type: replace-cross
Abstract: Textual data is often represented as realnumbered embeddings in NLP, particularly with the popularity of large language models (LLMs) and Embeddings as a Service (EaaS). However, storing sensitive information as embeddings can be vulnerable to security breaches, as research shows that text can be reconstructed from embeddings, even without knowledge of the underlying model. While defence mechanisms have been explored, these are exclusively focused on English, leaving other languages vulnerable to attacks. This work explores …

arxiv breaches can cs.ai cs.cl cs.cr data eaas information language language models large llms nlp research security security breaches sensitive sensitive information service text vulnerable