Beyond the benchmark: Advancing security at AI speed

Microsoft Security published a series of updates addressing AI-driven threats and agentic systems, plus active supply-chain and CI/CD risk. Key items: MDASH — an agentic vulnerability-detection system integrated across Windows, Azure, and identity; ASSERT — an open-source framework to convert specifications into executable agent evaluations; an updated taxonomy of agentic AI failure modes from a year of red‑teaming; guidance and telemetry playbooks for reconstructing AI activity in Microsoft 365 Copilot and Azure AI; research into a prompt‑injection pathway in Anthropic’s Claude Code GitHub‑to