AI RESEARCH
Beyond Indistinguishability: Measuring Extraction Risk in LLM APIs
arXiv CS.LG
•
ArXi:2604.18697v1 Announce Type: cross Indistinguishability properties such as differential privacy bounds or low empirically measured membership inference are widely treated as proxies to show a model is sufficiently protected against broader memorization risks. However, we show that indistinguishability properties are neither sufficient nor necessary for preventing data extraction in LLM APIs.