AI-generated editorial illustration; not a documentary photograph.
Cohesity introduced Cohesity Agent Resilience, a Data Cloud capability designed to back up enterprise AI agents’ memory and configuration and restore them to an earlier state after suspicious or damaging behavior. The product was announced at the company’s Catalyst conference, but access is currently restricted to select customers.
Recovery for Agent State
Administrators can select a recovery point from before an incident and restore only the affected memory or settings. Cohesity says the capability uses the immutable backup and clean-room recovery systems it already applies to on-premises, cloud and software-as-a-service workloads. It can also identify and protect databases, file systems and vector stores used by an agent, while a topology map shows connected systems, accessible data and potential protection gaps.
Limited Platform Support
The initial release covers agents built with Amazon Bedrock, including Bedrock AgentCore and Bedrock Agents. Cohesity expects general availability by the end of the year. Microsoft and Google agent platforms are on the roadmap, although the company did not provide release dates. This means organizations using other agent frameworks cannot yet apply the same recovery process through the service.
Cyber Resilience Automation
Cohesity also outlined Autonomous Cyber Resilience, a longer-term plan to use agentic workflows for data discovery, protection recommendations, threat scans, recovery rehearsals and incident response. Proposed actions would still require administrator approval. A related feature is available now for customers using Cohesity Data Cloud Enterprise Edition with Cohesity DSPM, allowing newly discovered sensitive data without existing protection to be added automatically to policy-based backup groups.
Survey Highlights Readiness Gap
The announcements accompanied Cohesity’s fifth annual Global Cyber Resilience Report. In a Vanson Bourne survey of more than 3,200 IT and security decision-makers across 11 countries, 55% said their organizations were unprepared to detect or contain unintended AI agent actions. Another 81% said their recovery plans would need moderate or significant changes to withstand cyberattacks accelerated by frontier AI models.
Source: Cohesity’s new Agent Resilience lets companies roll back AI agents that go wrong
SIGNAL TITANS
Leave a Reply