Cohesity has unveiled Agent Resilience, a new capability for its Data Cloud platform built to back up and restore enterprise AI agents and the systems they interact with.
Announced at Cohesity Catalyst, Agent Resilience addresses a growing problem as AI agents get greater access to databases, applications and other corporate resources. Security tools can identify suspicious agent behavior, but detection alone does not restore an agent or repair data that has been altered or deleted.
Agent Resilience protects two core elements of an AI agent: its memory and configuration. Cohesity uses point-in-time recovery so administrators can return an agent to a trusted state after an incident like memory corruption, a configuration problem or malicious activity.
The platform also protects resources used by agents, including databases and file systems. This makes a difference because an agent operating with privileged access may potentially make changes across multiple enterprise systems. Restoring the agent itself would not necessarily repair those downstream changes.
Cohesity is also providing topology information that maps an agent to its memory stores, applications, databases and supporting infrastructure. IT teams can use this view to identify dependencies and determine which resources need protection before an incident occurs.
The initial release supports AI agents built on Amazon Bedrock, including Bedrock AgentCore and Bedrock Agents. Cohesity plans support for Microsoft and Google agent platforms.
Not Ready for Rogue Agents
Companies are still developing their defenses against the operational risks created by AI agents. In Cohesity’s fifth annual Global Cyber Resilience Report, 55% of IT and security decision-makers said their organizations were unprepared to detect or contain unintended actions by AI agents.
The overall cybersecurity landscape is challenging. Some 69% of respondents said their organization experienced a material cyberattack during the previous 12 months, compared with 54% in the prior year’s report. Additionally, 81% said their recovery plans would require moderate or significant changes to withstand attacks accelerated by frontier AI models.
Beyond Agent Resilience, Cohesity outlined a longer-term strategy called Autonomous Cyber Resilience. The plan calls for using agentic workflows to automate portions of data protection, threat response, recovery testing and risk management.
Under the proposed model, IT teams could give Cohesity Copilot recovery requirements instead of configuring individual policies across applications. Data Cloud would then evaluate workloads and recommend protection settings, threat scans and recovery tests. Admins would need to approve actions before they are executed.
During an attack, automated workflows could assess affected resources, search for signs of attacker activity and prepare an isolated recovery environment. The strategy builds on Cohesity RecoveryAgent, which already automates parts of incident response and recovery.
Agent Resilience is currently available to select customers, with general availability targeted for the end of 2026.
