At its Catalyst conference, Cohesity introduced Agent Resilience, a new feature in its Data Cloud platform that protects and restores the memory and configurations of enterprise AI agents, allowing organizations to roll back agents to trusted states when errors or attacks occur.
- Backs up AI agent memory and configurations to enable rollback.
- Supports AWS Bedrock agents initially, with Microsoft and Google platforms planned.
- Part of broader automation plans for Autonomous Cyber Resilience by Cohesity.
What happened
Cohesity introduced a feature called Agent Resilience at its Catalyst conference. This new capability allows enterprises to back up the memory and configuration state of AI agents running in their environments. If an AI agent behaves unexpectedly or is compromised, administrators can roll back the agent to a previously known safe state, minimizing damage and downtime.
This function is designed to address a critical gap in current AI governance, since standard observability tools can detect anomalous behavior but cannot reverse changes made by an AI agent. Agent Resilience uses immutable backups and clean-room recovery methods already employed in Cohesity’s Data Cloud platform to protect AI agents’ state data.
Why it matters
AI agents in large organizations often act with extensive privileged access to sensitive data and systems, sometimes operating autonomously without continuous human oversight. This increases the risk of unintended or malicious actions that may corrupt or delete critical data. Existing governance tools can alert administrators to such risks but lack the ability to revert the AI agent's altered state or recover lost data.
By backing up AI agents' memory and configurations, Cohesity’s Agent Resilience provides a way to quickly restore AI systems to a trusted baseline without a full rebuild. This reduces downtime, operational disruption, and the risk of prolonged exposure to corrupted agent states, which could lead to severe security incidents or compliance breaches.
What to watch next
Currently, Agent Resilience supports AI agents built on Amazon Web Services’ Bedrock platform, including Bedrock AgentCore and Bedrock Agents. Cohesity plans to extend support to AI agents developed on Microsoft and Google platforms in the future. The feature is limited to select customers until general availability by year-end.
Cohesity is advancing a broader vision called Autonomous Cyber Resilience, which aims to automate cyber defense workflows. This includes natural language-driven policy recommendations and incident response orchestration via Cohesity Copilot and Maestro integrations with major AI assistants like ChatGPT and Claude. These efforts signal a push toward continuous, AI-enhanced cyber risk management at scale.