Extracting Training Data from Large Language Models

TL;DR

This paper shows that large language models can be exploited to extract individual training data, including sensitive information, raising privacy concerns and highlighting the need for safeguards.

Contribution

It introduces a training data extraction attack on large language models, demonstrating its effectiveness and analyzing factors influencing vulnerability.

Findings

Larger models are more vulnerable to data extraction attacks.

Hundreds of verbatim training examples, including PII, can be recovered.

Extraction is possible even when data appears in only one document.

Abstract

It has become common to publish large (billion parameter) language models that have been trained on private datasets. This paper demonstrates that in such settings, an adversary can perform a training data extraction attack to recover individual training examples by querying the language model. We demonstrate our attack on GPT-2, a language model trained on scrapes of the public Internet, and are able to extract hundreds of verbatim text sequences from the model's training data. These extracted examples include (public) personally identifiable information (names, phone numbers, and email addresses), IRC conversations, code, and 128-bit UUIDs. Our attack is possible even though each of the above sequences are included in just one document in the training data. We comprehensively evaluate our extraction attack to understand the factors that contribute to its success. Worryingly, we find that larger models are more vulnerable than smaller models. We conclude by drawing lessons and discussing possible safeguards for training large language models.