Snowflake adds agent spending limits and previews tool-call controls

Snowflake announced generally available controls that attribute AI spending to teams, agents or applications and enforce hard limits in near real time. Its generally available observability features record model calls, tool invocations and guardrail decisions in tables inside customers' accounts. The company also introduced public previews of Cortex AI Gateway Tools, which checks agent tool calls against policy using the end user's identity, and Cortex AI Guardrails, which adds runtime protection against prompt injection and jailbreaks. Code-execution sandboxes and automatic model routing remain in private preview.