AI Agents act at machine speed. Your resilience strategy should, too

By Ashish Gupta, Managing Director (India) and Head of Engineering, Rubrik India

For most of the past decade, enterprise AI governance centred on a reasonable anxiety: that machines might make mistakes humans hadn’t foreseen. Whether that be review cycles, approval workflows, or human checkpoints, the entire architecture of enterprise security rested on the assumption that a human is always in the loop.

That assumption is now broken. AI agents write code, commit changes, initiate recovery sequences, and move data through pipelines without pausing for human review. Enterprise security was engineered for human-paced operations, but AI agents don’t always comply. When something goes wrong, the damage accumulates in seconds.

Rubrik Zero Labs data shows 88% of enterprise leaders are concerned about meeting recovery time objectives (RTOs) as agentic threats increase. The data shows that this concern is not a forecast, but a real-time reflection of what security teams are experiencing. AI-driven threats compress the window between vulnerability discovery and active exploitation from months to seconds.

The harder problem is that organisations are being asked to secure environments they never designed for autonomous operation. Agents have been deployed into stacks never mapped for this kind of exposure.

The governance question business leaders are asking is “Who is accountable when an autonomous system makes a consequential decision?” The problem is that this question is being asked after deployment, not before.

When enterprises think about AI readiness, they focus on adoption. What gets far less attention is the infrastructure those deployments inherit.

The first gap is data visibility. Unstructured data accounts for roughly 90% of enterprise footprints, yet organisations have had to duplicate entire estates through costly Extract, Transform, and Load pipelines just to surface the less than 10% of data AI operations actually need. The result includes incurring expenses, creating compliance exposure, and stripping access controls in transit.

The second gap is recovery. Legacy backup tools restore individual resources, such as a file, a database, or a virtual machine. Modern cloud applications are interdependent stacks of code, configurations, secrets, and identities. Restoring a file when full application recovery is needed isn’t a solution. It’s a category error.

Third, enterprise security controls assume human review at every consequential step. When an agent can initiate dozens of actions between a vulnerability trigger and an analyst alert, perimeter-based controls become largely beside the point.

Visibility Before Governance
Much of the guidance circulating today tells organisations to “control” their AI. The problem is governance assumes visibility most organisations don’t yet have. You cannot govern agents you haven’t mapped, or recover applications whose dependencies you’ve never documented.

The more defensible sequence runs the other way: visibility first, governance second. That means knowing which agents are deployed, what each can access, and whether their actions have crossed a policy threshold, with the ability to reverse them precisely when needed. Without that foundation, security stays reactive. organisations that have invested in observability aren’t facing fewer threats. They simply have more options when threats materialize.

The most meaningful shift in enterprise security is a move from reactive planning to pre-emptive architecture, which includes mapping dependencies, validating clean restore points, and pre-building recovery sequences before conditions deteriorate.

The Gap Between Now and 2030
Gartner predicts that by 2030, 35% of organisations will utilize Cloud Application Infrastructure Recovery (CAIRS) solutions to complement infrastructure as code (IaC) disaster recovery orchestration, up from less than 5% in 2026. (Top Trends in Backup and Data Protection for 2026, Michael Hoeck, March 12, 2026). As a CAIRS, Autonomous Business Recovery for Cloud Applications provides a comprehensive, preemptive solution to recover the entire application environment.

The organisations that navigate this period most effectively won’t be those that deployed AI fastest.

They’ll be the ones that built the architecture to contain, audit, and reverse what autonomous systems do when they act in error. Over any meaningful time horizon, they inevitably will. Resilience, in the agentic era, is not an IT function. It is a strategic capability that sits with business leadership. The question now is whether leadership is building for it before an agent makes a decision no one anticipated.