Stack layer / Contrast
A single greedy decode manufactures self-improvement gains on a model that was never trained
arXiv 2608.20290
Stack layer / Contrast
Safety Nets Shrink Certifiable Aviation Neural Networks by Three Orders of Magnitude With 100% Correct Runtime Behavior
arXiv 2608.20053
Stack layer / Threat pattern
Secrets Sitting in an Agent's Context Leak Through Ordinary Replies, and Stronger Models Leak More
arXiv 2608.19857
Threat pattern / Contrast
OWASP Top 10 for LLM Applications barely agrees with 7,714 real incidents
arXiv
Policy dependency / Update thread
BreakGuard Catches 30.3% of Real Dependency Breaking Changes at About $0.90 Each
arXiv 2608.20167
Stack layer / Update thread
ShieldZFS Extends ZFS to Detect Rollback and Fork Attacks by a Malicious Cloud Provider on TEE Storage
arXiv 2608.19924
Contrast / Update thread
Humans Catch Partial Audio Deepfakes 9% of the Time and Call the Synthetic Sentence Real in 77% of Cases
arXiv 2608.19959
Policy dependency / Stack layer
Let the model pick NoThink, Short, or Long at response start and cut mean tokens 41% for a 1.4 point accuracy loss
arXiv 2608.20256