Rubrik Agent Identity, the company’s newest service announced at Black Hat in Las Vegas, aims to control artificial‑intelligence agents by granting access one tool call at a time.
How the service works
The offering builds on Rubrik Agent Cloud, a platform launched last October and made generally available for Anthropic’s Claude in June. Agent Identity focuses on the credential layer that sits beneath those deployments. Traditionally, agents inherit the broad, standing permissions assigned to human employees, leaving a single compromised or misbehaving agent able to modify SaaS applications, databases and APIs before any alarm sounds.
According to Rubrik, the new service monitors every agent and Model Context Protocol (MCP) server at runtime, controls access per tool call with fine‑tuned models, and remediates risky actions through identity checks, audit logs and the company’s Agent Rewind feature. Customers begin by inventorying active agents, MCP servers, skills and plugins, then limit access to specific servers and tools using an On‑Behalf‑Of federation that extends existing identity structures.
Enforcement occurs at an MCP gateway. Each tool call must clear three checkpoints: a behavioral analysis step where Rubrik’s SAGE governance engine evaluates intent, input parameters and likely impact; a policy check at the infrastructure layer; and an identity verification that authenticates the session and issues a scoped, short‑lived token. If a request falls outside an explicit policy scope, it is blocked before execution.
Related: Bending Spoons buys Airtable for 1 billion
Addressing a “shadow workforce”
Rubrik describes the unmonitored machine identities as a “shadow workforce.” Research from Rubrik Zero Labs indicates 86 % of global IT and security leaders expect AI agents to outpace existing security guardrails within a year, yet only 23 % claim full visibility into the agents already operating in their environments.
Dev Rishi, Rubrik’s general manager of AI, said, “Agents are no longer just synthesizing information, they are acting on the enterprise on behalf of employees, and the access models we built for humans and static credentials were never designed for autonomous actors.” He added that Agent Identity lets enterprises decide who can do what with agents and enforces it per tool call, at the moment of action, with scoped, short‑lived access and no standing permissions.
In practice, the approach mirrors earlier attempts to limit privileged access for human users, but it adds a layer of granularity suited to autonomous software. This shift resembles the move from static API keys to time‑bound tokens that many cloud providers adopted after several high‑profile breaches.
Agent Rewind, described as the only feature of its kind on the market, can reverse actions that clear the gateway yet still cause problems. Rubrik’s own survey found 88 % of leaders worry about meeting recovery‑time objectives as agentic threats scale, highlighting the need for such rollback capabilities.
Related: Artist Sues Over AI Generated Images
Integration and outlook
Agent Identity integrates with Okta and Microsoft Entra ID, extending enterprise identities to machine actors without requiring a separate directory. Rubrik has not disclosed pricing or a general availability date for the service.
The company, now positioning itself as “the Security and AI Operations Company,” entered the agent tooling space after acquiring Predibase in June 2025 for a deal valued between $100 million and $500 million. Rishi, who co‑founded Predibase and served as its chief executive, continues to steer Rubrik’s AI strategy.
Security remains a priority.
While the announcement marks a notable step toward tighter AI governance, the broader industry still faces challenges in tracking and mitigating the rapid proliferation of autonomous agents across diverse environments.