2 papers
cs.CR2026
Adversarial Entropy Inflation Against Gumbel-Based Inference Verification
Nikita Kezins
Gumbel-based inference verification bounds LLM weight exfiltration by only forgiving token choices that plausibly arise from honest GPU nondeterminism, reporting a >200x slowdown f…
cs.LG2026
Beyond Red-Teaming: Formal Guarantees of LLM Guardrail Classifiers
Nikita Kezins, Urbas Ekka, Pascal Berrang +1
Guardrail Classifiers defend production language models against harmful behavior, but although results seem promising in testing, they provide no formal guarantees. Providing forma…