Compound agents rising
GTC panels flagged “compound agents” — hybrid AI systems that stitch foundation models with domain intelligence to run multi‑day, multi‑step workflows and act as “coworkers, not just tools.” Boards are being asked to factor these new autonomous workflows into audit and risk frameworks because they change operational and data governance risk in real time. (youtube.com)
NVIDIA framed agent orchestration as the next enterprise interface at GTC 2026 and unveiled platform components including OpenClaw and Nemotron to support multi‑agent, multi‑step systems. (sdxcentral.com) (nvidia.com) Five security vendors announced governance integrations for NVIDIA’s agentic AI stack at GTC, with VentureBeat reporting four vendors in active deployments and one with a validated early integration. (venturebeat.com) VentureBeat’s coverage also cited survey data showing 48% of cybersecurity professionals rank agentic AI as the top attack vector heading into 2026 and only 29% of organizations feel fully ready to deploy these systems securely. (venturebeat.com) Audit functions face traceability shortfalls because agentic systems set sub‑goals, call external tools and persist memory across sessions, a pattern ISACA says weakens traditional audit trails and complicates post‑incident reconstruction. (isaca.org) Big‑four and advisory firms are publishing action frameworks for boards: KPMG’s Trusted AI guidance and EY’s multi‑layered agentic risk framework both prescribe continuous monitoring, embedded compliance, and live policy enforcement for autonomous workflows. (kpmg.com) (ey.com) Governance bodies and consultancies are urging specific board changes—Deloitte lists five board actions for AI oversight and NACD recommends institutionalizing governance that spans development, deployment and retirement of autonomous systems—shaping obvious director skill sets in audit, cybersecurity and AI policy. (deloitte.com) (nacdonline.org)
