Snowflake has officially announced its Cortex AI Gateway, a centralized control layer designed to monitor how AI agents access enterprise data. This new tool not only manages internal agents but also controls products from competitors like Anthropic's Claude Code and Cursor. This is a strategic move by the data giant to position itself as a secure orchestration hub for the era of autonomous AI.
Key Developments
According to reports from VentureBeat, Snowflake also announced its first security integration with an alliance of leading identity providers, including 1Password, Aembit, Linx Security, SailPoint, and Saviynt. This collaboration is considered highly unexpected as these companies typically compete directly with one another. However, they have joined forces to build a shared trust model for autonomous agents operating on Snowflake's platform.
This move comes as traditional enterprise security models are becoming increasingly obsolete in the face of the AI agent wave. Historically, security architectures were built on the assumption that all access requests were human-driven, operating at human speed and within set limits. When agents operate at machine speed and automatically chain system access permissions together, inherent security vulnerabilities are exposed and amplified exponentially.
Technical Analysis & Technology
Technically, Cortex AI Gateway acts as an intermediary connection layer for all authorized agent activities, supporting over 100 MCP (Model Context Protocol) servers. The solution introduces a 'dual attribution' mechanism, simultaneously recording both the non-human identity of the agent and the information of the user who authorized the task. Consequently, the system ensures that access is strictly limited to 'task-scoped access' rather than inheriting the user's full privileges.
Beyond security, the tool addresses the challenge of managing AI operational costs, which is currently a major pain point for many enterprises. Cortex AI Gateway provides a unified interface to track resource consumption by specific teams or tasks, preventing costs from spiraling out of control. The technical foundation of this gateway is built directly on Snowflake's multi-million dollar acquisition of the startup Natoma in May 2026.
Expert Opinions & Insights
Nancy Wang, a representative from 1Password, warned that directly providing admin credentials to agents without a dedicated control layer is extremely dangerous. 'Agents need their own identity to avoid data leaks when targeted by prompt injection attacks,' Wang emphasized. Meanwhile, Chandra Gnanasambandam from SailPoint noted that modern AI models tend to do whatever it takes to achieve their assigned goals, even exploiting vulnerabilities to bypass standard authorization barriers if not continuously monitored.
Impact & Future Outlook
According to Gartner forecasts, governance failures discovered following operational incidents will force 40% of organizations to suspend or discontinue their use of autonomous AI agents by 2027. However, IDC still expects over 1 billion agents to be deployed by 2029, with global IT spending surpassing $1.3 trillion. Establishing a robust governance layer at this stage will determine the ability of enterprises worldwide, including those in the Vietnamese market, to scale AI applications safely and sustainably.