Rubrik Develops AI to Judge Agent Actions

Data security firm Rubrik is developing a new AI system called SAGE (Semantic AI Governance Engine) designed to monitor and evaluate the actions of its AI agents in real-time.

The development addresses the growing need to manage and secure autonomous AI agents. Dev Rishi, Rubrik's GM of AI, presented the system at the VB Transform 2026 event. He noted that while organizations have created AI governance policies, enforcing and monitoring them in practice has proven challenging.

SAGE operates within Rubrik's Agent Cloud platform, replacing the traditional manual approval process. The system reads the semantic intent behind agent actions and compares it against policies written in natural language. Rishi described this as a shift from a 'human-in-the-loop' model to an 'AI-in-the-loop' approach, which initially raised caution among security leaders.

Rubrik's own research indicates that monitoring and approving AI agent actions often takes more time than the agents themselves save. SAGE aims to streamline this process and enable the full realization of AI's benefits. The system's economic efficiency is based on a small language model with significantly lower costs and latency compared to frontier LLMs.

Rishi emphasized that the primary challenge for companies seeking to achieve AI ROI is not cost or performance, but obtaining security and risk approval. SAGE aims to solve this by providing a reliable and automated oversight mechanism.