Snowflake Announces 'Cortex AI Gateway,' an AI Agent Management Platform

Snowflake announced 'Cortex AI Gateway,' a new capability for centrally managing AI agent access to enterprise data and tools. The feature is characterized by its ability to control access from external AI agents provided by competitors, including Anthropic's Claude Code and Cursor. The company simultaneously announced partnerships with five security firms: 1Password, Aembit, Linx Security, SailPoint, and Saviynt.

Until now, corporate security has been designed on the premise that 'humans are the ones using the systems.' Access rights have been managed under the assumption that users log in and operate one application at a time. However, AI agents move across multiple systems at speeds several times faster than humans and carry the risk of exercising permissions simultaneously that were never intended to be combined. Mayank Upadhyay, Chief Security and Trust Officer at Snowflake, commented on this challenge: 'AI is not creating entirely new security problems; rather, it is exposing blind spots that were previously invisible.'

In the early stages of AI agent adoption, risky operations were prevalent in the field. Nancy Wong, CTO of 1Password, points out that 'it became standard practice to give agents login credentials directly and let them act on your behalf.' What happens when agents operating with the broad permissions of an IT administrator autonomously move through an organization? It demonstrates that danger. Cortex AI Gateway provides a mechanism to monitor and control all agent actions in real time to address the risk of 'agents running amok as human proxies.'

The five companies listed as partnership partners are all firms with strengths in identity management—the mechanism that controls who accesses what and how. Typically competing companies have come together to build a common trust model for autonomous agents. Upadhyay stated, 'If each vendor builds a closed agent ecosystem, enterprises will recreate in the AI world the data silos they have spent years trying to eliminate,' emphasizing the importance of open interoperability.

This announcement can be positioned as a clear demonstration of Snowflake's intent to expand its role from a mere 'data storage location' to a 'control platform' that determines what AI agents are permitted to do. As enterprises seek large-scale deployment of AI agents, cost inflation and unexpected data access have emerged as management challenges, and demand for management tools is growing. The fact that five identity management competitors are participating in a common framework suggests that industry-wide efforts to standardize security for agents are beginning.

When agent AI becomes fully operational within enterprises, the traditional authentication model of 'confirming trust once at login' will become insufficient. As Upadhyay states, 'Trust must be continuously verified across all agents, all actions, and all operations,' necessitating a fundamental rethinking of access management approaches. Whether Cortex AI Gateway functions as a solution will depend on future enterprise deployment cases and the actual depth of collaboration with partner companies.