Cohesity Launches Agent Recovery Tool for Enterprise AI

Cohesity has introduced a new capability called Agent Resilience, designed to protect and recover the infrastructure behind enterprise AI agents. The tool addresses a critical gap in current security measures: while existing systems can detect when an AI agent behaves unexpectedly, they generally cannot undo the damage if the agent corrupts or deletes data. By leveraging snapshot architecture and immutable backups, the new feature allows organizations to restore agents and their associated resources to a known-good state.

The launch marks a shift in how companies approach AI governance. As AI agents increasingly take autonomous actions like updating workflows and accessing privileged credentials, the risk of unintended consequences grows. Cohesity’s solution aims to automate parts of the cyber resilience process, reducing the manual effort required to respond to and recover from AI-related incidents. The company announced these developments at its Catalyst event in Singapore, alongside a free training course for attendees.

Current security tools lack recovery power

A significant portion of enterprises are struggling to manage the risks posed by automated AI workflows. According to the fifth annual Cohesity Global Cyber Resilience Report, 56% of organizations stated they are not well prepared to detect or contain unintended actions by AI agents. Furthermore, 58% expressed low confidence in their ability to verify the integrity of AI models after a cyberattack. This highlights a disconnect between monitoring capabilities and actual recovery options, leaving many businesses vulnerable to data loss caused by autonomous systems.

Vasu Murthy, Chief Product Officer at Cohesity, emphasized that detection alone is insufficient. He noted that while tools can identify when an agent goes off course, they cannot reverse the changes it has already made. Agent Resilience is built to provide a precise recovery path for both the agents themselves and the databases or file systems they interact with. This approach treats AI infrastructure as a critical asset requiring the same level of protection as traditional enterprise data.

Initial support limited to Amazon platforms

At launch, Cohesity Agent Resilience supports Amazon Bedrock AgentCore and Amazon Bedrock Agents. The company has indicated that support for Microsoft and Google agent platforms is on the roadmap, but these are not yet available. General availability for all supported services is targeted for the end of 2026. This phased rollout means that enterprises using other major cloud providers must wait for full compatibility, which could limit the immediate utility of the tool for multi-cloud strategies.

The feature works by protecting agent memory and configuration using point-in-time recovery capabilities. This allows teams to roll back to a stable state following memory corruption, misconfiguration, or malicious activity. It also extends protection to the external services agents manage, ensuring that affected resources can be precisely recovered. While this provides a strong safety net, the initial limitation to Amazon's ecosystem is a significant trade-off for organizations relying on diverse cloud infrastructures.

Free academy course offered to attendees

Alongside the technical launch, Cohesity introduced the AI Resilience Academy, starting with a free, self-paced course titled Foundations of AI Resilience. Attendees at the Catalyst event received early access to this resource. The course is designed to help organizations manage risk as AI systems evolve from advisory tools to active participants in business processes. By providing educational resources, Cohesity aims to equip IT teams with the knowledge needed to implement effective resilience strategies alongside the new software.

The company also outlined its vision for Autonomous Cyber Resilience, which seeks to automate a five-step framework using agentic workflows. This includes protecting data and identities, ensuring recoverability, remediating threats, practicing recovery, and optimizing risk posture. By integrating automation into these steps, Cohesity hopes to reduce the manual workload associated with incident response. This broader vision positions the company at the forefront of AI-driven security, though the full implementation of these automated workflows remains a future goal.