The company’s research found that 56 % of organisations do not feel well prepared to recover from unintended actions by AI agents, while 58 % lack confidence in verifying AI model integrity after a cyber‑attack. In Australia, 47 % said they have no centralized inventory of AI agents, copilots and related workflows.
Agent Resilience is designed to safeguard the memory, configuration and connected resources of AI agents, including databases, file systems and other services they can modify. The service uses snapshots, immutable backups and clean‑room recovery – techniques already employed by Cohesity for other workloads across on‑premises, cloud and SaaS environments.
At launch, the feature supports Amazon Bedrock AgentCore and Amazon Bedrock Agents, with support for Microsoft and Google agent platforms slated for later in the year. It also provides an “agent topology” view that gives administrators a single pane of glass showing each agent, its memory stores, linked applications and infrastructure, helping teams identify dependencies for rapid restoration.
Vasu Murthy, Cohesity’s chief product officer, said recovery remains a weak point in many AI oversight programmes, noting that detection alone cannot undo changes made by a rogue agent. He described Agent Resilience as a “precise recovery path” for both agents and the resources they impact, and linked it to the company’s broader “Autonomous Cyber Resilience” strategy.
The Autonomous Cyber Resilience model seeks to automate Cohesity’s five‑step cyber‑resilience framework – protection, recoverability, remediation, recovery rehearsal and risk‑posture improvement – using agentic workflows. Administrators would set business recovery objectives in Cohesity Copilot, which would then continuously discover assets, assess protection status, recommend policy changes and orchestrate incident response, with human approval for key actions.
Cognizant’s Srikanth Kuntamukkala, CIS Global Lead for Partner Sales, welcomed the development, saying that as AI agents take on more business‑critical tasks, the ability to restore an agent’s state and the data it altered will become essential for managing risk in production environments.
Alongside the product launch, Cohesity introduced an AI Resilience Academy, offering a free, self‑paced course that distinguishes between AI assistants that make recommendations and AI agents that can act autonomously. The training aims to help organisations understand the new risk vectors that emerge when AI systems move from advisory roles to making live changes.