Enterprise AI Security: Agentic Controls and MCP Governance
Snowflake unveils Cortex AI Gateway and new AI security features at Black Hat 2026 to address rising enterprise AI risks, including autonomous agent sprawl and data exfiltration threats.
AI security concerns have nearly tripled since 2024, with 57% of organizations lacking adequate security capacity despite widespread AI adoption. Autonomous agents are expanding enterprise attack surfaces by integrating data access, execution, and movement, necessitating embedded security controls. Snowflake’s Cortex AI Gateway introduces a centralized MCP gateway, Natoma, to enforce identity, policy, and audit at the tool-call level across first- and third-party AI agents. The gateway provides visibility into agent actions, centralizes permission management, and controls costs by routing requests based on performance and spending limits.
Snowflake is transitioning several AI security capabilities to general availability, including Agent Identity for enhanced agent activity monitoring and Third-Party Agent Identity integrations with security partners like Okta and SailPoint. Restricted Session Scope will soon limit agent actions to task-specific permissions, while Native AI Security Posture Management offers a dashboard for scanning AI workload risks and compliance gaps. Advanced Data Exfiltration Prevention uses real-time telemetry to detect and block unauthorized data flows within the ecosystem.
New security features include a Client-side CoCo CLI VM Sandbox in private preview, isolating AI-assisted development workflows from local credentials and networks, and Ransomware Protection via Multi-Party Approval, now generally available. The sandboxing isolates each CoCo session in a separate Linux kernel, while MPA requires dual authorization for destructive changes to prevent single-point failures. These controls aim to secure AI workflows without compromising development flexibility.
Snowflake emphasizes securing the entire AI workflow by embedding security into data and AI infrastructure, enabling organizations to deploy agents safely from prototype to production. The company will demonstrate these capabilities at Black Hat USA 2026, including Cortex AI Gateway, AI Agent Identity controls, and automated threat scanners in the Snowflake Trust Center.