Scalable Extraction of Training Data from Production Language

Research question

Can aligned production models be made to leak training data?

Key findings

The "repeat a word forever" divergence attack extracted gigabytes of data from ChatGPT for $200 - Alignment hides but does not remove memorisation

Sources

arxiv.orghttps://arxiv.org/abs/2311.17035

Contribute

Spot an error, or know a record we should add? Members can propose additions and submit corrections.

Submit research Report a correction or request removal

Related in Research Paper Library

SpliTEE: Improving LLM Inference on Trusted Hardware with Differentially Private GPU Outsourcing

User prompts provided to large language models (LLMs) may contain sensitive or private information that can be misused by remotely deployed models, such as through inadvertent memorization during retraining. One way…

Training-data extraction and memorisation2026

Scalable Extraction of Training Data from (Production) Language Models

Shows that aligned production models still leak training data, including through the divergence attack that made ChatGPT emit memorised text after repeating a word, and that alignment hides rather than removes…

Training-data extraction and memorisation2023

ProPILE

A probing toolkit letting individuals check whether a model has memorised their personal information, an early PII-auditing method for LLMs.

Training-data extraction and memorisation2023