Rubrik (NYSE: RBRK) announced the release of its Semantic AI Governance Engine (SAGE), which the company describes as the data security industry's first AI governance engine designed to control autonomous agents in real time.

SAGE powers Rubrik Agent Cloud and uses a custom Small Language Model (SLM) to interpret policy meanings and provide oversight for enterprise AI agent operations. The system translates natural language instructions into machine logic and identifies potential policy violations before they occur.

The governance engine includes semantic policy interpretation, a proprietary SLM, adaptive policy improvement capabilities, and integrated remediation features. When agent errors occur, SAGE triggers Rubrik Agent Rewind to reverse destructive actions and restore data integrity.

Rubrik conducted benchmark testing comparing its custom SLM against OpenAI's GPT-5.2. According to the company, its SLM processed messages five times faster and achieved higher accuracy in detecting policy violations while reducing compute overhead associated with real-time AI monitoring.

"SAGE marks a pivotal moment in AI security as we shift the focus from if agents can be deployed to how they can be governed at scale," said Devvret Rishi, General Manager AI at Rubrik.

The company positioned the product as addressing governance bottlenecks in enterprise AI deployment, where traditional systems rely on deterministic rules that cannot adapt to dynamic agent actions or comprehend natural language.

Rubrik included a safe harbor statement noting that unreleased services referenced may not be made generally available on time or at all, and customers should base purchase decisions on currently available services.