Snowflake has announced Cortex AI Gateway and a set of AI security updates aimed at giving enterprises a central way to govern AI agents and track associated usage costs as “agentic” AI tools spread across applications and platforms.
The company said Cortex AI Gateway is designed to address two barriers to enterprise AI adoption: securing AI agents and improving visibility and control over AI consumption costs. The gateway is positioned as a control layer for both Snowflake-native agents and third-party agents built on external platforms, including Claude Code and Cursor.
Snowflake said the service will provide governance over how agents access models, tools, Model Context Protocol (MCP) servers and enterprise systems, with capabilities including request routing and cost optimisation across models and workloads. The company said this is intended to give organisations more visibility into agent activity and token usage, and to help prevent unexpected cost growth.
The announcement builds on Snowflake’s acquisition of Natoma in May 2026, which Snowflake said brought enterprise MCP platform capabilities into its platform to support agent interoperability. Snowflake also announced integrations intended to support third-party agent governance and auditing, naming 1Password, Aembit, Linx Security, Okta, SailPoint and Saviynt as initial partners.
In a statement, Snowflake Chief Security and Trust Officer Mayank Upadhyay said security needs to be central as enterprise AI shifts toward agent interoperability and that organisations require visibility, governance and control when agents from different platforms access data and invoke tools.
Snowflake said the gateway is intended to help address security and governance challenges associated with AI agents collaborating across enterprise environments, where traditional security architectures may not provide centralised oversight of cross-system activity.
The company also outlined broader AI security work it said is aimed at improving AI risk posture visibility, providing agents with verified identities, protecting sensitive data, applying context-aware controls and limiting agent sessions to task scope. Snowflake cited BlackRock and Thomson Reuters as customers enhancing security using its zero-trust approach for agentic AI.
Thomson Reuters Head of Data & Analytics Caitlin Halferty said strong security, governance and visibility are needed to build “trusted AI” for professionals, and that enterprises need controls that protect sensitive information while supporting innovation as agentic AI moves into production.
Snowflake said some related capabilities are generally available now, while others are in private preview or expected to enter public preview soon. The company said Cortex AI Gateway will be in public preview soon, and that several partner integrations will be in private preview soon, with the Okta integration planned for private preview in Q4 2026.
