Commvault adds agent recovery for Bedrock & Foundry

Commvault has introduced AgentRecover features for AI agents built on Amazon Bedrock AgentCore and Microsoft Foundry.

The additions are intended to help organisations audit agent activity and recover affected systems across connected environments.

The new functions address a growing challenge for IT and security teams as AI agents gain access to data, applications and infrastructure. In that setting, an unintended action by an agent can affect multiple systems in quick succession, with consequences for data integrity, application availability and day-to-day operations.

AgentRecover extends Commvault's recovery tools to the agent platform layer through three main functions. It can discover supported agents and map the data, identities, configurations, applications and infrastructure they use; audit what an agent did and what it may have changed; and restore both the agent and the underlying systems affected by its actions.

Customers can also use the product alongside Commvault's identity recovery and change-auditing tools to examine which identities an agent acted through and what data was changed.

Why it matters

The announcement reflects a wider shift in enterprise software as business processes increasingly move from conventional applications into AI-driven agents. That changes how resilience and recovery are handled because the scope of an incident may extend well beyond a single workload or data set.

Amazon Bedrock AgentCore and Microsoft Foundry are used to build, deploy and operate AI agents for business processes. By adding support for those platforms, Commvault is seeking to cover both the agents themselves and the systems they interact with.

Pranay Ahlawat, Chief Technology and AI Officer at Commvault, outlined the company's rationale for the move.

"Business logic is moving out of applications and into agents, and agents don't behave deterministically. A single agent decision can trigger changes across data, applications, and infrastructure on multiple platforms," said Pranay Ahlawat, Chief Technology and AI Officer at Commvault.

"Enterprises need to see and audit what agents have done, then recover both the agent and affected resources. That's what we're delivering for customers building on Amazon Bedrock AgentCore and Microsoft Foundry," Ahlawat added.

Operational risk

The issue is particularly sensitive in regulated sectors where AI systems may interact with customer records and transaction workflows. In those cases, recovery may need to address not just a model or agent configuration but also the knock-on effects on the systems touched by that agent.

"As AI agents take on a greater role in business, resilience has to extend beyond the agent itself," said Michele Buschman, Chief Information Officer at American Pacific Mortgage.

"Agents may interact with regulated loan data, applications, and customer-facing systems. Recovering both the agent and affected systems after an unintended action would be essential for mortgage companies focused on compliance, keeping loans on track, and maintaining customer trust," Buschman said.

Industry analysts are also focusing on how recovery methods need to adapt as agentic systems become more common in enterprises.

"Agentic AI changes the resilience equation. Traditional recovery assumes a discrete set of data at a point in time," said Phil Goodwin, Research Vice President at IDC.

"In contrast, agentic recovery must recover the agent, data, and resources - all at machine speed. Commvault's move to audit agent activity and recover the agentic resources reflects a rapidly growing requirement for enterprise resilience," Goodwin said.

Platform support

The initial support covers Amazon Bedrock AgentCore and Microsoft Foundry, two environments used by organisations building production AI agents for business tasks. The aim is to give teams a way to identify agent dependencies, understand what an agent changed, and focus recovery efforts on those changes rather than unnecessarily reverting entire environments.

That approach could appeal to enterprises trying to contain the operational fallout from unwanted actions without causing broader disruption to systems that were not affected. It also places AI agents within the same recovery planning discussion as applications, data stores, identities and infrastructure, rather than treating them as isolated tools.

As more companies test or deploy AI agents in operational settings, the question for IT departments is shifting from whether agents can complete tasks to how organisations can trace, contain and reverse the effects when those tasks go wrong.