Saviynt launches Snowflake integration for AI agent risk

Saviynt has launched an integration with Snowflake aimed at reducing identity risk from third-party AI agents.

The integration is designed to help organisations connect, govern and audit AI agents operating across business systems and platforms. It sits within Snowflake's security ecosystem and focuses on access controls for autonomous software agents that interact with enterprise data and tools.

The announcement comes as companies face new security questions from the growing use of AI agents in business workflows. Unlike conventional software automation, these agents can make autonomous decisions, access sensitive information and trigger actions in other systems. That raises concerns about how to verify their permissions and monitor their behaviour.

Traditional identity and access management tools were largely built around human users and, in some cases, service accounts. As a result, security teams are considering how to adapt governance models for software agents that may act across several platforms, invoke external tools and operate with a degree of independence.

Saviynt said the integration uses its Access Gateway product to assess what an agent is trying to do, the context of the request, the permissions assigned to the agent and the risk involved at runtime. The aim is to ensure agents are authorised only for the specific task being performed and to help detect privilege misuse or unusual activity before it affects business operations.

Akhil Kapoor, Senior Vice President, Business Development, Saviynt, said the issue reflects a gap in current enterprise security frameworks.

"Traditional identity security models were built for human users, not AI agents, creating a critical gap in enterprise security. Our integration with Snowflake addresses this challenge through Saviynt's Access Gateway, which evaluates agent intent, context, permissions, and risk at runtime to ensure AI agents are only authorised to perform the actions required for a specific task. It prevents privilege misuse and anomalous activity before it can impact the business," Kapoor said.

Industry shift

The companies cast the partnership as part of a broader shift in cybersecurity towards governance frameworks for what is often described as the agentic enterprise. That term refers to operating environments in which AI agents handle tasks across multiple systems with limited human intervention.

For businesses adopting such tools, the challenge is not only whether an agent can gain access to a system, but whether it can be restricted to approved actions, observed in real time and audited after the event. These issues are drawing increasing attention as enterprises look to deploy AI models and agents in customer service, software development, data analysis and back-office processes.

Snowflake, which markets its platform as an AI data cloud, said trusted interoperability between agents will require a different approach to security from existing models. It framed the integration as part of building that foundation.

"Securing the agentic enterprise requires an entirely new model of trust, visibility, and control," said Mayank Upadhyay, Chief Security and Trust Officer, Snowflake. "Through our integration with Saviynt, we are establishing a secure foundation for agent interoperability, equipping organisations with the precise access capabilities and trust required to run autonomous workflows safely."

Access controls

The focus on agent interoperability reflects a broader trend in enterprise technology, where businesses want AI systems from different vendors to exchange information and carry out tasks across data, applications and infrastructure. That may improve efficiency, but it also expands identity risk because a third-party agent may need permission to access internal systems, data repositories or external services.

Security specialists have increasingly argued that identity is becoming a central control point for AI governance. In practice, that means treating non-human actors such as bots, applications and AI agents as identities that require authentication, authorisation and continuous oversight, rather than assuming existing user-based rules will be sufficient.

In this integration, Saviynt's approach is to apply those principles to third-party agents operating through Snowflake's environment. The aim is to give enterprises a way to connect agents across platforms while maintaining clearer governance and audit trails over what those agents are permitted to do.

The launch underlines how software suppliers are repositioning identity security tools for the rise of autonomous AI in business operations. As organisations move from experimenting with AI assistants to deploying agents that can take actions on their behalf, the question of how those agents are governed is becoming a more immediate operational issue.