Varonis Atlas Adds Intent-Based Access Controls for AI Agents
Summary
Varonis says its Atlas Agent Intent-Based Access Control evaluates AI-agent prompts, reasoning, tool calls and session context against intended purpose, with configurable alerts, blocking, approvals, quarantine and audit trails.
Key points
- Varonis says Agent IBAC compares an agent’s instructions with its reasoning, selected tools and tool parameters to identify actions that drift from the requested purpose.
- The system evaluates prompts, responses and tool calls across a session to identify cumulative drift and multi-turn jailbreak attempts.
- Organizations can tune detection sensitivity and define an Agent Manifest covering an agent’s purpose, ownership, permitted data and prohibited actions.
- Configurable runtime responses include alerting, blocking, modifying or logging actions, and requesting human approval.
- Atlas can quarantine an identity or session for a customer-defined period, with administrators able to lift, extend or make the quarantine permanent.
- Varonis says Atlas records prompts, responses, tool executions and guardrail actions in an audit trail.
Article Details
- Event Type
- Product capability announcement: Varonis describes Agent Intent-Based Access Control in Varonis Atlas.
- Impact
- Varonis says Agent Intent-Based Access Control evaluates AI-agent actions against their instructions and can alert, block, modify, log, or route actions for approval. It can also quarantine an identity or session and provide an audit trail. The article describes these as product capabilities, not as responses to a reported incident.