Tag
This paper introduces natural identifiers (NIDs) for post-hoc privacy auditing and dataset inference in large language models, eliminating the need for retraining or held-out datasets.
This paper identifies distribution shift and scale constraints as critical failure modes for statistical contamination detection methods in LLM benchmark auditing. Evaluating three paradigms across 27 models reveals only 199 correct outcomes out of 335 evaluations, indicating a systematic reliability gap that prevents these methods from replacing transparent data provenance.