Varonis Introduces Agent IBAC for AI Agent Security in Atlas Platform

Varonis Atlas Enhances AI Security with Agent IBAC

Varonis has announced Agent Intent-Based Access Control (IBAC), a new feature integrated into its Varonis Atlas platform. This capability aims to secure AI agents by ensuring they operate within their intended boundaries when interacting with enterprise data. The introduction of Agent IBAC comes in response to concerns about AI agents potentially misusing access, such as exposing sensitive data or performing destructive actions.

Real-time Enforcement and Anomaly Detection

Agent IBAC functions by comparing an AI agent's received instructions with its reasoning and the tools or data it attempts to access. If a discrepancy is detected, the system can respond in real time by alerting administrators or blocking the action. For significant deviations, Atlas can quarantine the AI agent's identity and block subsequent actions for a specified period.

Granular Control Based on Impact

The system allows for configurable responses based on the potential impact of an agent's deviation. For instance, a clear mismatch between an instruction (e.g., checking weather) and an action (e.g., invoking a migration tool) would result in automatic blocking. In cases where an agent's action drifts but poses no data risk (e.g., setting a reminder instead of a one-time answer), Agent IBAC can simply log the deviation without interrupting the process.

Addressing Limitations of Traditional Access Control

Traditional role-based access control systems were not designed to evaluate the contextual actions of non-human identities like AI agents. These static controls can be circumvented by agents that elevate privileges or access data outside their intended scope. Agent IBAC addresses this by enforcing permissions at runtime, focusing on whether an agent should be allowed to act on specific data within a given context, rather than just whether it has general access.