Observability for AI Systems: Strengthening visibility for proactive risk detection

Microsoft Security Blog items (Mar 2026) emphasize AI security and observability, governance innovations, and active threat campaigns. Key themes: need for enhanced observability and monitoring of AI/agentic systems to detect prompt injection and other AI-specific risks; new Purview features for governance and safe AI adoption; detection and playbooks for prompt abuse; Microsoft DART case study on Teams voice-phishing enabling identity-led intrusions; active campaigns (Storm-2561) using SEO poisoning to distribute signed trojans via fake VPN clients to steal VPN credentials; the Contagious “/”