Validate the Statistical Independence Assumption in Multi-Sample Fingerprinting

Validate whether the noise affecting repeated three-step inference-engine fingerprinting probes is statistically independent, particularly when multiple samples are drawn from a long-running inference-engine instance in a self-refinement scenario.

Background

The paper proposes repeating a three-step fingerprinting workflow to reduce stochastic errors and estimates the number of probes needed to identify an inference engine with 95% confidence. That estimate assumes that the noise affecting each probe is statistically independent.

The authors note that this assumption may fail in self-refinement settings because repeated samples from a persistent engine instance can experience confounding effects. Establishing whether independence is justified is necessary to assess the reported sample-complexity and confidence estimates.

References

That being said, the assumption of statistically-independent noise is a non-trivial one that requires more validation by future work.

— Inference-Engine Fingerprinting Attacks are Practical: Exploring Model-Driven Environmental Discovery, Exploitation, and Escape  (2609.20614 - Radway et al., 17 Sep 2026) in Section 5.2, “Fingerprinting Results,” paragraph “Multi-sample fingerprints”