Saviynt adds AI agent runtime controls & verification

Saviynt has added new controls to its Agent Access Gateway for AI agents, introducing Intent-Aware Runtime Authorization, or IARA.

The feature assesses AI agent actions in real time against identity, context, policy and intent. It can block actions that fall outside approved limits and create an audit event.

The release reflects a broader shift in how companies are trying to govern AI systems as software agents move from pilot projects into day-to-day business use. Unlike conventional software accounts, these agents can make decisions, adapt to changing conditions and carry out large numbers of actions across corporate systems in a short period.

That creates a problem for security teams that have historically relied on access controls built for human users and fixed applications. In that model, a user or service is granted permission in advance, but the system may not judge whether a specific action is appropriate at the moment it is attempted.

Saviynt is positioning Agent Access Gateway as a runtime authorisation layer to address that gap. The system is designed to govern what AI agents can do as they interact with applications, data, APIs, tools, infrastructure and other agents.

The new IARA function adds another check by examining not only whether an agent has technical permission, but also whether the action matches the reason the agent is acting. Saviynt says this allows organisations to stop unintended or unauthorised actions before they are completed.

One example from Saviynt involves a sales operations agent that is allowed to access customer relationship management data to summarise a sales opportunity. If that same agent then tries to export customer records, alter pricing terms or trigger outbound communications, the question becomes whether those actions align with the original request, not whether the software could technically perform them.

In that scenario, intent-aware checks are meant to identify the mismatch at runtime and prevent the action. The approach reflects a growing focus in enterprise security on continuous verification rather than one-time permission grants.

Alongside IARA, Saviynt expanded features in its Identity Security for AI offering. These include controls that let organisations evaluate AI agent actions at runtime, apply static and dynamic policies to tool access, and determine whether an agent is acting independently, on behalf of a human, or through another AI agent.

The company also introduced inbound and outbound access controls for AI agents. Those controls are intended to govern who is allowed to interact with AI agents and which applications, data and other resources the agents may access.

Such measures are aimed at extending identity governance practices already used for staff, contractors and software accounts to autonomous systems. For companies adopting AI agents more widely, the challenge is not only to manage what the agents can reach, but also to maintain accountability for actions taken across chains of users, tools and systems.

Native integrations have also been extended to include Microsoft Foundry, N8N and Snowflake Cortex, among others. The additions suggest Saviynt is seeking to place its controls closer to the environments where organisations are building and deploying AI workflows.

Verification focus

Separately, Saviynt added identity verification features intended to reduce impersonation risk. The issue has gained attention as AI tools make fraud, social engineering and unauthorised access attempts easier to carry out.

The verification functions support human identity certification within the Saviynt platform and include biometric scanning, selfie photos, liveness detection, and support for more than 4,000 government-issued document formats across more than 177 countries.

The combination of agent governance and identity verification shows how identity security vendors are responding to two related pressures from AI adoption: controlling autonomous software actors inside enterprise systems, and strengthening trust in the humans who initiate, approve or review sensitive actions.

Vibhuti Sinha, Chief Product Officer at Saviynt, framed AI agents as a new category of identity inside large organisations.

"AI agents are becoming a new class of enterprise identity - autonomous, powerful, and capable of taking action across critical business systems," said Vibhuti Sinha, Chief Product Officer at Saviynt.

He said the new controls are designed to move access decisions closer to the moment when actions occur.

"Agent Access Gateway gives enterprises a way to control AI agent behaviour at runtime, when decisions actually happen. With IARA, organisations can move beyond static permissions and make access decisions based on what an agent is trying to do, why it is doing it, and whether that action should be allowed," Sinha said.

The latest changes broaden Saviynt's effort to build identity and governance controls around AI agents across development environments, runtime systems and enterprise applications where those agents take action.