Cohesity unveiled Cohesity Agent Resilience, which backs up enterprise AI agent memory and settings and rolls them back to a trusted point in time if problems arise.

SiliconANGLE reported on Sept. 16 that Cohesity showcased Cohesity Agent Resilience at its Catalyst conference.

Cohesity said Cohesity Agent Resilience targets AI agents inside companies that query databases and carry out tasks with high privileges. Tools to detect anomalous activity already exist, it said, but there is a lack of ways to recover content that agents have changed, corrupted or deleted, prompting it to introduce Agent Resilience.

Vasu Murthy (바수 머시), Cohesity's chief product officer, said, "Detection can tell you that an AI agent has deviated from its path, but it cannot roll back changes."

Agent Resilience first protects memory and settings based on Cohesity's immutable backup and recovery capabilities. If a problem occurs, an administrator can choose a recovery point from before suspicious activity began and restore only the affected memory or settings. It also provides a function to locate and protect the databases, file systems and vector stores used by an agent. It also offers a topology map showing the systems connected to each agent and the data it can access.

Support is currently limited to AWS Bedrock-based agents. It includes Bedrock AgentCore and Bedrock Agents. Microsoft and Google agent platforms are expected to be supported in the future.

Cohesity also unveiled an "autonomous cyber resilience" plan. If an administrator presents recovery time objectives and recovery point objectives to the natural-language assistant Cohesity Copilot, Cohesity Data Cloud evaluates workloads and recommends protection policies, threat scans and recovery rehearsals.

Execution does not take place without approval. Customers using Cohesity Data Cloud Enterprise Edition and Cohesity DSPM can automatically protect data that has not yet been protected among newly discovered sensitive data.