Snowflake Launches AI Gateway, Security Suite for Enterprise Agents

At Black Hat 2026, Snowflake announced Cortex AI Gateway, a centralized control and governance layer designed to manage the growing complexity and risk of autonomous agent deployments across enterprise systems. The gateway integrates Natoma, an MCP (Model Context Protocol) gateway, to enforce identity, policy and audit controls at the point where AI agents call tools—addressing a hard problem enterprises now face: fragmented agent access, invisible tool usage, and exploding AI costs.

Cortex AI Gateway provides three core capabilities. Control allows teams to grant, restrict and audit model and tool access from a single endpoint instead of manually configuring each new agent type, supporting fine-grained authorization across 100+ MCP servers. Visibility captures agent actions in real time—which tool was called, which system it touched, in what order and by whom—feeding comprehensive audit trails for security and compliance. Cost and performance includes automatic model routing based on cost, latency, capability and data residency, plus budget guardrails by team, agent or workload. Several features—Wide Model Catalog (bringing GPT, Gemini, Claude, Grok, Mistral, GLM and others under unified governance), Access Governance and Sprawl Control, Observability and Tracing, AI Cost Control, and Intelligent Model Routing—are in private preview.

Snowflake complemented this new gateway with a suite of AI security capabilities moving into general availability and public preview. Agent Identity (GA) enables governance teams to enforce data access policies specific to agent sessions, preventing sensitive data access even when agents run on behalf of privileged users; third-party agent identity integrations with security vendors (1Password, Aembit, Cyera, Linx Security, Okta, SailPoint, Saviynt) extend the same policies to external AI tools. Restricted Session Scope (GA soon) limits what an agent can do to only what the task requires, ensuring read-only analysis stays read-only. Native AI Security Posture Management (GA) now fully integrates proactive risk scanning, compliance assessment and programmatic remediation into the Snowflake Trust Center. Ransomware Protection via Multi-Party Approval (GA) requires two or more authorizations before destructive system changes, removing single points of failure from sensitive architecture. Advanced Data Exfiltration Prevention (preview) pairs real-time telemetry with strict data movement policies to detect and block unauthorized sensitive data fetches triggered by agents, unauthorized data routing and mass downloads. Client-side CoCo CLI VM Sandbox (private preview) isolates AI-assisted development workflows in separate Linux kernels on macOS, minimizing credential and local storage exposure to AI workloads.

The announcement arrived as AI security concerns surged from 17% in 2024 to 48% in 2026, according to The Linux Foundation's 2026 State of Tech Talent Report. While 97% of organizations report commitment to implementing AI, 57% face significant capacity gaps in security and risk management. Autonomous agents have accelerated this pressure by expanding the attack surface—combining data access, system execution and data movement creates new vectors for tool hijacking, data exfiltration and cost runaway that legacy monitoring tools and application-layer patches cannot address. Snowflake positioned Cortex AI Gateway as foundational: by moving security controls into the data and AI infrastructure planes themselves rather than bolting them on afterward, enterprises can scale agent deployments without sacrificing visibility, governance or cost control. Demonstrations and product team details are available at the Snowflake booth (#8206) at Black Hat USA 2026.