Rubrik Inc. today turned its data security platform into an autonomous agent and made its control layer for Anthropic PBC’s Claude generally available, the headline items in a wave of announcements at its Rubrik Forward conference in Las Vegas.

The releases share a common pitch: as enterprises hand more work to AI agents, the question that matters is how fast an organization can recover when an agent, or an attacker abusing one, does something destructive. Rubrik, which now calls itself the “Security and AI Operations Company,” is answering it with agents of its own.

The centerpiece is Rubrik AI, which the company said gives its platform an agent-first interface across both Rubrik Security Cloud and Rubrik Agent Cloud. Customers define a business outcome and the software reasons over their data, identities and deployed agents to carry it out. The design is patent-pending. It runs in an agentic mode built on a single agent spanning the two product suites, applies guardrails meant to keep every autonomous action auditable, attributable and reversible, and runs multi-step recovery sequences that the company said previously took human teams weeks.

“Today, Rubrik becomes an agent,” said co-founder, Chairman and Chief Executive Bipul Sinha. “Our Agentic Cyber Resilience is designed to mitigate risks from both external AI attacks and internal agent deployments. Rubrik AI will deliver agentic automation to enable resilience against machine-speed cyber breaches and compromised AI agents.”

Separately, Rubrik moved its Agent Cloud for Claude Code and Claude Cowork to general availability. The service builds on the broader Agent Cloud Rubrik launched in October and targets a specific gap: AI agents now write, push and deploy code on their own, while most enterprise security was built assuming a human stays in the loop. That gap, the company said, opens the door to rogue commits, repository ransomware, prompt injection and intellectual property theft at machine speed.

The Claude version centers on a handful of capabilities. A governance engine the company calls SAGE, short for Semantic AI Governance Engine, applies intent-driven controls to autonomous agents in real time in place of static manual oversight. An agent inventory maps risk, access permissions and policy violations across deployed agents. Agent Rewind, which Rubrik bills as the industry’s only such feature, reverses unintended actions from any agent, including those built in Claude Code and Cowork.

Two features go further into developer workflows. Codebase Resilience keeps continuous immutable snapshots of GitHub and Azure DevOps repositories, stored outside the repo. If an agent or an attacker force-pushes over commit history or deletes every branch, Rubrik can roll the code back to a known-good state. The company said that covers ransomware rollback for code. A second layer backs up and version-tracks the configuration that governs how Claude agents behave, including system prompts, tool permissions, skills and files such as CLAUDE.md, then monitors for drift and flags changes that look malicious before they spread.

“Organizations are adopting Claude faster than any agentic technology we have seen and every security leader asks the same question: how do we stay in control when an agent can act?” said Anneka Gupta, chief product officer at Rubrik. “Rubrik Agent Cloud gives organizations a resilience layer for Claude, which allows them to see what agents can access, govern what they do, rewind their actions, and recover both the code and the agent’s own configuration when something is destroyed or tampered with.”

To get the Claude service in front of large enterprises, Rubrik also announced Project Hourglass, a partner program under which six global systems integrators will resell and embed Agent Cloud for Claude Code. Cognizant Technology Solutions Corp., Deloitte & Touche LLP, LTM Ltd., HCL Technologies Ltd., NTT Data Inc. and Wipro Ltd. are the launch partners, with Rubrik offering them joint go-to-market support, certification tracks and access to its engineering teams. Several said they are folding the service into their own AI platforms and frameworks, including Cognizant’s Neuro AI platform, Deloitte’s Ascend delivery platform and HCLTech’s VERITY Frontier AI Resilience framework.

The agent push leans on Rubrik’s own research arm for urgency. The company cited data from Rubrik Zero Labs showing 86% of cybersecurity leaders expect AI agents to outpace their security guardrails within the next year and a separate finding that 88% of leaders are concerned about meeting recovery time objectives as agentic threats increase.

Rubrik used Forward to extend its recovery story to cloud applications as well. Autonomous Business Recovery for Cloud Applications, now in private preview, is built on what the company calls its Preemptive Recovery Engine and aims to rebuild an organization’s “minimum viable business” after an attack rather than restore individual resources one at a time. The system maps an application and its dependencies, validates clean recovery points and pre-builds recovery plans during normal operation, then sequences a rebuild of network, compute and data when an incident hits. Rubrik pointed to a Gartner forecast that 35% of organizations will use cloud application infrastructure recovery tools by 2030, up from less than 5% this year.

“Recovering data isn’t recovering a business. If the application isn’t running, the restore fails,” Gupta said. “Machine speed attacks have made the old all-or-nothing recovery untenable.”

The company also pushed into AI data pipelines with the next stage of Rubrik Annapurna, which it positions as an AI-ready layer for unstructured data. Annapurna scans and indexes billions of files across network-attached storage and object stores in place, then publishes a queryable catalog into a lakehouse so data teams pull only the subsets they need for training and inference. The approach is meant to avoid duplicating an entire data estate and the extract, transform and load work that has kept most enterprise files out of AI systems. Unstructured data accounts for about 90% of a typical enterprise footprint, the company said. Annapurna is available now for qualified enterprise partners, with native lakehouse connectors promised in future releases.

Availability varies across the lineup. Agent Cloud for Claude Code is generally available, Annapurna is open to qualified partners and Autonomous Business Recovery is in private preview. Rubrik noted in a safe harbor statement that referenced features may not ship on schedule or at all.

The announcements follow a first-quarter revenue beat and raised outlook that Rubrik reported last week. The company expanded agent governance work with Google LLC at Google Cloud Next in April and partnered with CrowdStrike Holdings Inc. on identity resilience last year.

Photo: Rubrik