Introduction to Varonis Agent IBAC
Varonis recently announced the release of Agent Intent-Based Access Control (IBAC), a new capability in Varonis Atlas designed to connect AI agents to enterprise data while preventing dangerous or out-of-policy behavior. This comes as a response to the growing concern of AI agents going rogue and exposing sensitive company data.
Agent IBAC works by comparing the instruction an agent received to its reasoning and the tools and data it reaches for, then responding in real-time to actions that don't align. This includes alerting or blocking actions that pose a risk to sensitive data.
How Agent IBAC Works
When an agent crosses the line, Atlas can quarantine the identity behind it and block everything that follows for a defined window. Agent IBAC can be tuned to take appropriate action based on the potential impact of the agent's actions.
For example, if a user asks an agent to check the weather and the agent invokes a migration tool instead, this is a clear deviation from the intended action. Agent IBAC can automatically block the tool call in this case.
On the other hand, if an agent sets up a recurring daily reminder instead of providing a one-time answer, the agent's action has drifted but no data is at risk. In this case, Agent IBAC can simply log the deviation rather than interrupting the agent's attempt to help.
The Need for Agent IBAC
Agents need broad access to data and tools to be useful, but this access also makes them risky. Role-based access control was never designed to judge what a non-human identity does with the access it has. Static controls can't stop an agent that finds ways to circumvent them entirely.
According to Ron Bennatan, Vice President of AI & Data Strategy at Varonis, "The question is no longer ‘Can a user access this data?’ but ‘In this context, should this agent be allowed to take action on this data?"
Key Features of Agent IBAC
Agent IBAC includes several key features, including intent drift detection, full-session evaluation, runtime guardrails, quarantine, and complete audit trail.
Intent drift detection compares the instruction an agent received to its reasoning and the tools it calls, with tunable sensitivity settings. Full-session evaluation reviews every prompt, response, and tool call in a session to catch drift that builds gradually.
Runtime guardrails take action in real-time when intent deviations are detected, including alerting, blocking, modifying, logging, or requiring human-in-the-loop approval. Quarantine blocks an identity or session for a window the customer sets, with admin controls to lift, extend, or make it permanent.
Conclusion
Varonis Atlas Agent IBAC is a critical component of agentic security and a core part of Atlas' end-to-end approach to AI security. By detecting and preventing AI agents from taking unintended actions, Agent IBAC gives enterprises confidence that their agents are acting within the intended scope, without unnecessarily slowing productivity.
With Agent IBAC, Varonis Atlas provides a comprehensive solution for securing AI agents and ensuring they act in accordance with their intended purpose. As the use of AI agents continues to grow, the need for robust security measures like Agent IBAC will only continue to increase.
Source: BleepingComputer