Snowflake Cortex AI Gateway: the AI agent control that enterprises needed

What happened?

Snowflake on Tuesday unveiled Cortex AI Gateway, a centralized control layer designed to govern how AI agents—including those built by competitors like Anthropic's Claude Code and Cursor—access enterprise data, tools, and models. Alongside the gateway, the company announced a first wave of security integrations with 1Password, Aembit, Linx Security, SailPoint, and Saviynt, an unusual coalition of identity providers that often compete with each other, now aligned around a shared trust model for autonomous agents.

Why it matters

The announcement, made from Snowflake's unofficial headquarters in Bozeman, Montana, represents the company's most aggressive move to position itself not just as the place where enterprise data lives, but as the control plane that decides what AI agents can do with it. According to Mayank Upadhyay, Snowflake's chief security and trust officer, "the next era of AI will not be built with more walled gardens. It will be built through secure agent interoperability."

The core argument is that decades of enterprise security architecture are based on an assumption that no longer holds: that the actor behind every access request is a person. "Traditional security was built for a world where humans were the actors. AI agents change that completely," Upadhyay explained. Agents operating at machine speed can "combine access across systems and act on permissions that were never intended to be exercised together, amplifying long-standing risks."

Nancy Wang, CTO of 1Password, described the failure mode in visceral terms: "Imagine you're the security chief and you have admin access to all systems. Well, now your agent has admin access to all systems, so it could exfiltrate data if it falls victim to a prompt injection."

What impact will it have?

Cortex AI Gateway allows enterprises to define granular access policies for agents, audit their actions in real time, and prevent runaway costs by limiting which models and APIs they can consume. This addresses a growing problem: AI inference costs can skyrocket if agents make millions of unauthorized calls. Additionally, by integrating with identity providers, it creates an ecosystem where trust is continuously verified, not just at login.

For IT and security departments, this means they can adopt AI agents without losing control. For startups building agents, like Anthropic or Cursor, it implies they will need to comply with Snowflake's standards if they want to operate in enterprise environments. In the long run, Snowflake could become the trust intermediary for agent-to-agent interaction, similar to what Okta is for human authentication.

What should readers know?

• Not a standalone product: Cortex AI Gateway integrates with the Snowflake ecosystem (Cortex AI, Snowpark, etc.) and third-party tools.
• Continuous trust model: Unlike traditional authentication, the gateway verifies every agent action in real time.
• Competitors are welcome too: Snowflake allows agents from Anthropic, Cursor, and others to be governed by its gateway, aiming to become the de facto standard.
• Cost focus: In addition to security, the gateway controls API and model consumption to prevent cost leaks.
• Availability: The product is in preview; general availability is expected in the coming months.

"The challenge is not that AI creates completely new security problems. It's that AI exposes the blind spots we've always had," Upadhyay said.

Analysis

Snowflake's move is strategic: instead of competing directly with agent builders, it positions itself as the governance layer enterprises need to adopt agents securely. This echoes how databases became data platforms, or how firewalls became essential for network security. If successful, Snowflake will not just be a data warehouse, but the operating system for enterprise agents.

However, success will depend on adoption by agent providers and Snowflake's ability to maintain customer trust. The coalition with identity providers is a smart step, but fragmentation in the agent market could make standardization difficult.