AI Runtime Security & Agentic Zero Trust

Dominant story: Black Hat 2026 revealed 21 issues/12 CVEs in AI agent frameworks. 14% attack surface increase from AI, 90% shadow AI, 43-day patch time. 58% incident rate among enterprises that enforce without isolation. New architectures (per-agent identity, mTLS, human-in-the-loop) emerging. Multiple vendor moves (Lineation.ai, Neo, Cloudflare, Zscaler) and validation gaps highlighted. SOC automation reliance reinforces human-in-the-loop challenges. New: A 90-day CISO implementation plan for agentic AI security provides concrete steps beyond checklists, covering discovery, machine identity baselines, behavioral monitoring, and kill switches. RAG pipeline security labs also highlight poison retrieval and tenant isolation risks.