What happened
At Snowflake Expedition 2026 (Nov 3), Snowflake announced new agent-governance capabilities beyond the Cortex AI Gateway GA: Cortex Sandboxes (private preview) give agents a secure, private space to run custom code alongside enterprise data while inheriting the running user's permissions, shut down automatically, and avoid moving data off Snowflake; MCP governance through Cortex AI Gateway Tools (public preview) authenticates every agent tool call with the end user's identity and validates against fine-grained authorization policies; Business Ontology and Cortex Sense ground agents in approved business definitions.
Why it matters
These close two gaps in agent security: safe execution of agent-generated code inside the data platform, and authenticated, policy-checked MCP tool calls that prevent identity spoofing and over-broad tool access. It also extends governance/observability to all agent actions.
Applicability
Snowflake customers running agents against enterprise data, especially those with MCP tool integrations, should evaluate Cortex Sandboxes and MCP governance as they move to preview/GA.