What if the number codes AI uses to “understand” your text… could be reversed?
A 2023 study showed embeddings can leak the exact sentence, even personal info.
Think vault? It's invisible ink.
🔗 substack.com/@harrisrache...
🔬https://arxiv.org/abs/2310.06816 #AIprivacy #VectorDB #RAG #AIsecurity
arxiv.org
Text Embeddings Reveal (Almost) As Much As Text
How much private information do text embeddings reveal about the original text? We investigate the problem of embedding \textit{inversion}, reconstructing the full text represented in dense text embed...
