# An AI now judges every move Rubrik's agents make, its AI chief says

But no one has measured whether this AI judge is actually correct.

- **Source:** [VentureBeat](https://venturebeat.com)
- **Published:** 2026-07-23
- **Topic:** [Data & AI](https://brifing.si/c/data-ai)
- **Original article:** <https://venturebeat.com/security/an-ai-now-judges-every-move-rubriks-agents-make-its-ai-chief-said-at-vb-transform-2026-but-no-ones-measured-if-the-judge-is-right>
- **On Brifing:** https://brifing.si/s/4f4654d8-91bf-4441-a060-0f27e4040824

## Introduction

Rubrik's GM of AI Dev Rishi revealed at VB Transform 2026 that the company has developed a system where an AI evaluates every action of its AI agents in real time. However, the accuracy of this system remains unmeasured.

## Security policies vs implementation

At a CISO roundtable, all participants had written AI governance policies but struggled with practical implementation. Rubrik's 'YOLO mode' experiment exemplifies this dilemma in action.

## The YOLO mode experiment

Rubrik configured its data security agents to operate without human approval. Each action is evaluated by an AI called SAGE against company policies. This radical approach is being tested internally first.

## Developer pushback

The requirement for every command to run in approval mode during Claude Code and Cowork pilots sparked significant developer resistance. A 120-message Slack thread criticized it as 'security theater.'

## The efficiency paradox

Rubrik Zero Labs research found monitoring and approving agent actions takes more time than the agents save. A survey of 1,600+ IT leaders supports this counterintuitive finding.

## SAGE's role

The Semantic AI Governance Engine (SAGE) serves as an arbitration layer in Rubrik Agent Cloud, monitoring every agent action and interpreting the intent behind it against natural language policies.

## The security bottleneck

Rishi's 200 customer conversations revealed security approvals - not cost - as the primary barrier to AI ROI. While 66% of enterprises allow or are moving toward zero-human-review deployments, only 5% fully trust automated evaluations.

## Limits of written rules

Rubrik's internal policies exposed enforcement challenges. A rule like 'agents must respect customer data policy' requires interpreting three pages of legal text - something traditional systems can't do.

## Language models as auditors

Rishi's team bet that language-understanding models could police other models, leading to SAGE. This approach provides intent-reading capabilities beyond traditional tools.

## Conclusion

Rubrik's experiment establishes a new paradigm in AI governance. Systems like SAGE promise automated oversight based on semantic understanding rather than rigid rules. But their reliability remains unproven.

## Sources covering this story

- https://venturebeat.com/security/an-ai-now-judges-every-move-rubriks-agents-make-its-ai-chief-said-at-vb-transform-2026-but-no-ones-measured-if-the-judge-is-right — 

---

_Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._