Signal Agentic AI Security: Snowflake's Data-Model-Agent Framework / HPE brings agentic AI into production with NVIDIA

Summary

Snowflake published a security framework for agentic AI organized into three layers: a data layer enforcing least privilege, masking, data-movement controls, sovereignty and compliance; a model layer protecting the intelligence engine from manipulation within the customer's security boundary; and an agent layer governing agent behaviour, tools, identity, approvals and auditability. Snowflake argues agentic AI cannot be secured by any single feature and requires defence in depth, and introduced 'Agent Identity' (public preview) to tag every action an AI agent takes with a distinct, auditable signal in near real time. Around the same period, at HPE Discover Las Vegas 2026, HPE announced it is bringing agentic AI into production with NVIDIA, integrating the NVIDIA Agent Toolkit — including Nemotron open models, NemoClaw and the OpenShell secure runtime — into an agent operating system that reasons, monitors behaviour, enforces policy and reduces deployment risk. Rollout is staged: HPE Data Fabric Software arrives in October 2026, while additional Private Cloud AI features including agentic observability and NVIDIA Confidential Computing arrive in Q4 2026. Both efforts reflect a broader 2026 industry shift toward hardening the infrastructure and governance layers needed to move agentic AI from experimentation into enterprise production.