Cohesity Launches Recovery Tools for AI Agent Infrastructure

Cohesity has introduced a new capability called Agent Resilience, designed to safeguard the internal infrastructure of enterprise AI agents. This feature focuses on preserving the memory and configuration states that drive autonomous workflows, ensuring that organizations can restore these digital workers to a stable state if they malfunction or are compromised.

The launch marks a shift in how data security firms approach the risks associated with agentic AI. By leveraging existing snapshot and backup architectures, the company aims to bridge the gap between detecting an agent's erratic behavior and actually reversing the damage it may have caused to connected systems.

Securing Autonomous Workflows

The new tool specifically targets the state of AI agents, including their memory and settings. It utilizes point-in-time recovery mechanisms to allow teams to roll back an agent to a known-good version. This capability is crucial because while detection systems can alert users that an agent has deviated from its expected path, they cannot inherently undo the changes that agent has already implemented.

The protection extends beyond the agent itself to the resources it manages, such as databases and file systems. This ensures that if an agent corrupts data or misconfigures services, the organization can precisely recover those specific affected resources without needing to rebuild entire environments from scratch.

Vision for Automated Response

Cohesity is also outlining a broader strategy for autonomous cyber resilience. This approach involves using agentic workflows to automate the entire cycle of protecting data, identifying threats, and remediating issues. The goal is to reduce the manual workload on security teams by allowing them to define high-level objectives through a copilot interface rather than manually configuring complex policies across different applications.

This vision builds upon existing orchestration tools that help coordinate incident response. By connecting protection, response, and recovery capabilities, the company intends to create a unified plan for IT and security teams. This integration is designed to streamline operations, ensuring that defensive measures are applied consistently across the entire digital landscape.

Availability and Trade-offs

According to GN auto tech/cloud, the capability is currently available to select customers, with general availability expected by the end of 2026. At launch, it supports Amazon Bedrock, with plans to extend compatibility to Microsoft and Google platforms. The primary trade-off for early adopters is the limited scope of integration, as the system relies on specific cloud architectures and does not yet cover all emerging AI frameworks.

Additionally, the company launched a free, self-paced learning path to help teams understand these concepts. While this resource aids in grasping the theoretical underpinnings of agent resilience, it does not replace the need for practical configuration expertise. Organizations must still navigate the complexities of implementing these recovery protocols within their own diverse IT environments.