Snowflake expands CoCo AI guardrails to costs, access, and external tools

Snowflake's July announcement of CoCo governance centered on three ideas: cost control, grounding AI in enterprise context, and bringing trusted AI to where work happens. This follow-up builds specifically on cost and access governance, recognizing a core tension in enterprise AI adoption: teams want to unlock developer productivity but security and platform teams need to limit risk exposure. Before these controls, the only way to manage both was to restrict CoCo's availability—confining it to sandboxes, limited user groups, or manual approval workflows—which defeated much of the productivity promise.

The new layered approach addresses this by making governance visible and queryable rather than reactive. Per-user quotas provide spend boundaries that update daily, preventing surprise line-item costs and removing the need for manual tracking. Organization-wide policy (MDM), team-level agent profiles, and restricted session scope (RSS) each operate at a different scope and are enforced before a session starts, not audited afterward. Together with the Cortex AI Gateway—which brings managed, approved access to external systems like Jira, Slack, and Salesforce—administrators now have concrete, auditable answers to the security questions that have historically blocked broader CoCo rollout. The setup is designed to be invisible to developers in the normal case, surfacing only when a boundary is about to be crossed.