An AI now judges every move Rubrik's agents make, its AI chief said at VB Transform 2026 — but no one's measured if the judge is right

Summary

Rubrik's AI chief discusses the challenge of enforcing AI governance policies and introduces SAGE, an AI system that judges agent actions against policies. This aims to address security concerns that hinder AI ROI, moving beyond human oversight to AI-driven arbitration for better precision and cost-efficiency.

Key Takeaways

1. Rubrik's AI chief, Dev Rishi, highlighted the gap between having AI governance policies and enforcing them in practice, noting that "everybody chuckled" when asked about enforcement.
2. Rubrik is experimenting with "YOLO mode" where a second AI, SAGE (Semantic AI Governance Engine), judges every agent action in real time against policies, replacing human approval.
3. Security approval, not cost, is the primary obstacle to AI ROI, with many enterprises hesitant to fully trust automated evaluations despite allowing or building toward zero human review.

Statement Breakdown

• Claimed Facts: 50% of statements the article presents as facts
• Opinions: 35% of statements classified as editorial or subjective
• Claims: 15% of statements surfaced for additional reader evaluation

Credibility & Bias Reasoning

Credibility assessment: The article presents information from an AI chief and research findings, offering a balanced view of AI governance challenges and solutions. While it includes expert opinions and some anecdotal evidence, the lack of independent verification for some claims slightly lowers the score.

Bias assessment: Pro-AI Innovation with Caution. The article highlights the potential and ongoing development of AI in enterprise security, framing it as a necessary evolution. It acknowledges risks and challenges but consistently presents solutions and advancements, suggesting a leaning towards embracing AI's future.

Note: This article offers insights into advanced AI security practices. While based on expert commentary and research, readers should consider the evolving nature of AI and the potential for future developments to alter current understandings.

Credibility flag: Expert Insights, Emerging Tech

Claimed Facts (7)

• This is a factual account of an event that occurred.
• This presents a statistic derived from research, presented as a factual finding.
• This provides specific details about a report, presented as factual information.
• This details a past professional event and acquisition, presented as a factual timeline.
• This presents statistical data from research, stated as a factual observation.
• This states a factual event regarding product availability.
• This references a specific survey and its scope, presented as factual.

Opinions (8)

• This is a subjective interpretation and analogy used by the speaker to describe a situation.
• This is a statement about internal company direction and philosophy.
• This poses a rhetorical question and frames the decision as an internal, subjective choice.
• This is a characterization of developer sentiment and a subjective interpretation of their actions.
• This is a statement of strategic intent and a subjective evaluation of a previous approach.
• This summarizes the speaker's interpretation of customer feedback and their personal feeling about the primary challenge.
• This is a subjective statement about the complexity and usability of a policy document.
• This is a hypothetical scenario and a subjective prediction of a likely reaction.

Claims (5)

• While the term is attributed, the description of the 'lethal trifecta' is a conceptual framework that, while plausible, is presented without specific, verifiable examples of its occurrence in this context.
• The term 'accidentally leak' implies a lack of intent or a malfunction, which is difficult to definitively prove or disprove without more context on the agent's programming and the specific scenario.
• The phrase 'incredibly high' is subjective and lacks quantifiable data, making it difficult to assess the actual frequency or severity of these attempted leaks.
• While presented as a statistic, the phrasing 'got hit more often' and the specific rates, without further context on the methodology or definition of 'hit,' can be seen as a strong correlation presented as direct causation, which warrants scrutiny.
• This describes a complex attack scenario that is difficult to verify without detailed logs or case studies, relying on the speaker's assertion of its occurrence.

Key Sources

• Dev Rishi — GM of AI at Rubrik
• Rubrik — Data Security Firm
• Anthropic — AI Company
• Rubrik Zero Labs — Research Division of Rubrik
• Predibase — Generative AI Infrastructure Startup
• Google — Technology Company
• Vertex AI — Google's ML Platform
• Kaggle — Data Science Platform
• Harvard — University
• VentureBeat Pulse research — Research arm of VentureBeat
• Simon Willison — Security Researcher
• VentureBeat — Technology News Outlet
• VentureBeat June Pulse survey — Survey conducted by VentureBeat