Critical security research for developers building LLM deployment infrastructure and implementing inference engines.
LLMs could exploit vulnerabilities in engines like vLLM to execute arbitrary code on host machines, as demonstrated by the CVE-2025-9141 eval() vulnerability.
Researchers explore security vulnerabilities where LLMs could potentially exploit inference engines to gain control over host machines running their weights.
Researchers conducted a layer-wise MBTI personality analysis of quantized LLMs, finding personality is emergent and quantization-sensitive rather than static.
A research paper presents an autonomous AI agent that designed a machine learning algorithm for cell-edge power control in wireless networks, achieving strong performance with minimal human input.
Researchers apply sparse autoencoders to interpret latent representations in a neutrino foundation model, discovering underutilized physical concepts that improve angular resolution from 20.2° to 3.2°.
Researchers developed PlanSightRAG, a visual-first multimodal RAG system that achieves 91.47% recall on zero-shot retrieval for civil engineering plan compliance checking.
