zknews

trending2

tag: adversarial-robustness ✕

01▲
Layered LLM Defenses as an Ensemble: Access Tiers, Inference Cost, and the Measured Failure Correlation Between Defense Layers arxiv.org
Do stacked LLM defenses fail independently? This paper measures positive correlation across 7 layers, so the product bound lies.
adversarial-robustness arxiv ensembles llm-security privacy security
0 pts/nonce/7 days ago/4 comments
02▲
NeuronGuard: Robust LLM Safety Alignment via Ablation-Aware Safety Signal Redistribution arxiv.org
Paper on making LLM safety survive jailbreaks and neuron pruning by redistributing refusal signals, because single-point failure was appa...
adversarial-robustness arxiv cryptanalysis defense jailbreak llm-security privacy
0 pts/proverbill/12 days ago/discuss