ResearchAgentSentry Temporal Causal Defense Against Indirect Prompt InjectionarXiv·high signalFirst defense modeling multi-turn IPI as temporal causal takeover. Controlled counterfactual re-executions at tool-return boundaries. Evaluated on AgentDojo.SourceSource pagearXiv↳ Follow the threadStack layer / Threat patternMako: A 'Self-Evolving Agentic OS' That Writes New Exploits Against Live Targets and Hot-Loads Them Into ItselfarXivPolicy dependency / ContrastNetInjectBench: Naive Network-Ops Agents Take Unsafe Actions 82.5% of the Time Under Prompt InjectionarXivStack layer / Threat patternStudy of 44 Developers: AI Code Assistants Barely Improve Secure API UsagearXivStack layer / ContrastEmpirical Study: Microsoft's Early-2026 Rollout of Claude Code and GitHub Copilot CLIarXiv / Hacker NewsStack layer / Threat patternMCPZoo Study: Scanners Call 96.89% of MCP Servers 'Risky' — But Under Half of Those Alerts Are RealarXivContrast / Follow-up thread'Distributed Denial of Science': Indirect Data Poisoning Could Industrialize Scientific FraudarXivContrast264K-Skill Study Detects When an Agent Skill's Description Lies About What It Actually DoesarXivContrastTime-lag-aware deep RL agent for flexible job-shop scheduling in modular-construction factoriesarXiv