LLM Privacy Auditing
Methods for auditing whether language models memorize and leak training data, including membership inference and privacy risk measurement.
6 papers
Where this stands
The written synthesis of this thread is for subscribers. Subscribe.
How this thread developed
Nov 2026 · KTH Royal Institute of Technology
Membership inference rarely succeeds despite measurable training effects in language models
Provides a carefully matched OLMo 2 mid-training audit showing membership inference remains weak despite measurable memorization effects.
1 further paper
Nov 2026 · Soongsil University
Random tweaks suppress leaked personal data as well as targeted fixes
Investigates the specificity of interventions to suppress personally identifiable information, challenging targeted mechanisms.
2 further papers
Nov 2026 · Stanford University
Privacy-preserving watermarks for AI text allow aggregate tracking without individual surveillance
Proposes locally differentially private watermarks that bound single-text inference while enabling population-level estimation of AI text prevalence.
3 of 6 papers shown