Stack layer / Update thread
SpecGuard Turns Speculative Decoding's Acceptance Rate Into a Free Runtime Backdoor Detector
arXiv
Stack layer / Update thread
NCP-ArchPreview reaches OLMo-3-7B's final pretraining loss on 51.3% of the tokens by predicting multi-token concepts
arXiv / HuggingFace Daily Papers
Stack layer / Contrast
A2ABreak extracts a 37-state machine from the A2A spec and finds 11 protocol-level flaws that a fully compliant attacker can exploit
arXiv
Stack layer / Update thread
100 LLM agents running a simulated town economy for 26 weeks froze the money supply: 0.3% of prices ever changed, and memory deletion made no difference
arXiv
Stack layer / Contrast
Skill optimization via contextual bandits cut optimization cost 55-58% using only 50 examples per benchmark
arXiv 2609.11682
Stack layer
py-kvcache Loads KV Cache From NVMe 2x Faster Than LMCache at 80k Tokens, but Loses on an H100
arXiv
Policy dependency
Tencent study: agents cross authorization boundaries in 55-62% of runs when a control constraint is missing and an unsafe action is executable, and 87% when compaction drops the constraint
arXiv
Contrast
T1, a 122B MoE Terminal Agent, Goes From 43.8% to 64.0% on Terminal-Bench 2.1 With RL on a Real Shell
arXiv