ResearchCUDA Agent Agentic RL for CUDA Kernel GenerationarXiv·high signalAgentic RL system outperforms torch.compile by 100% and beats Claude Opus 4.5 by 40% on GPU kernelsSourceSource pagearXiv↳ Follow the threadStack layer / Threat patternPiSAs Benchmarks Cross-User Privacy Leaks in Shared Agentic SystemsarXivStack layer / Update threadVisual Action Outcome Reasoning Alignment Improves VLM Physical ReasoningarXivStack layerVera Framework Auto-Generates Agent Safety Tests; 93.9% Attack Success on Production FrameworksarXivStack layerCoding Before Testing Biases an LLM's Own Test Suite Toward Its Buggy CodearXivStack layerSemi-formal reasoning: force the agent to state premises and trace execution before it verifies a patcharXivThreat patternLook-Ahead Bias Reframed as a Verifiable Non-Interference Property for Agentic TradingarXivStack layer / Threat patternArmadin details full sandbox escape in Anthropic's Claude Cowork; Anthropic disputes, silently patchesSiliconANGLEStack layer / Threat patternAirflowAttack Fools Infrared Remote-Sensing VLMs With Physical PerturbationsarXiv