Protect and secure your data from cyber attacks
Data Protection
Data Security
Data Insights
The 5 Steps to Cyber Resilience
Cloud & SaaS
Enterprise
Industries
Backups, agents, sovereignty, and the Minimum Viable Company—what Cohesity is talking about at Booth #1874 this year.
Something changed at AWS re:Invent this year that’s easy to miss. AWS merged re:Inforce, its standalone security conference, into the main event. Security used to get its own week in June. Now it shares the stage with all the innovation from AWS. This move makes sense given the onslaught of cyberattacks we're seeing these days.
If you’ve moved critical workloads to AWS, you’ve probably read the shared responsibility model at least once. This delineation has shaped security in the cloud for over a decade. AWS runs the infrastructure and runs it well. What happens to your data, your identities, and your ability to come back after someone encrypts production and wipes your backups on the same afternoon is your problem. Between geopolitical tension, attackers armed with advanced AI, and agents acting inside your environment, that risk is real.
That’s the territory we’ll be covering at Booth #1874 all week. Here’s what we want to talk with you about at AWS re:Invent.
Most teams we talk to understand the importance of immutable backups and protecting all their critical workloads on AWS. Good. The problem shows up two steps later, when you restore a workload and discover the attacker is still in it. Dormant malware, a persistence mechanism nobody found, a service account that got compromised three weeks before anyone noticed. You bring the system back, and the attacker immediately reinfects your environment. Incident responders have a name for this, the doom loop, and it’s the difference between a bad weekend and a bad month.
We recently published a reference architecture for Cohesity Data Cloud on AWS that’s built around not falling into that loop. It has five layers: immutable backups across data center and cloud, anomaly detection on the backup data itself (encryption spikes, odd deletion patterns, the things that mean something got past your EDR), a cyber vault with its own identity and access domain, a clean room on AWS built from dedicated accounts and isolated VPCs, and a validation pipeline that scans and tests each workload before it’s allowed anywhere near production.
The idea underneath all of it is simple: a restored workload is untrusted until you’ve proven it isn’t. The nice part about doing this on AWS is that the Clean Room only exists when you need it. You spin it up sized to the incident instead of paying for a second data center that sits idle.
The 3-2-1 rule is a tried-and-true maxim for IT. Three copies, two kinds of media, one offsite. It’s a good rule. It’s outlasted tape, disk, and the first decade of cloud because the things it protects against haven’t changed: hardware fails, people delete things, and buildings flood.
What it never had to think about was an adversary. A flood doesn’t read your runbook. An attacker does. They map your environment, find the backup servers by name, pull the admin credentials, and delete the backups before they touch production. If your three copies sit on one network under one identity system, you don’t have three layers. You have three targets.
That’s why Cohesity has long advocated for the 3-2-1-1 rule. The extra one is a copy that’s isolated, immutable, governed by its own credentials, and verified clean before you trust it. In practice that’s a cyber vault. It stays disconnected most of the time. The air gap opens long enough to write or restore, then closes. Quorum approvals mean not even a single stolen password gets you in.
The vault matters for a second reason that doesn’t get enough attention. After an attack, the most recent backup is the one most likely to be infected. You need a copy the attacker couldn’t have touched, and you need somewhere to inspect it. In our 2026 Global Cyber Resilience Report, 60% of organizations hit by a material attack said they lost time because they couldn’t confirm the restored data was clean. This represents a significant trust gap in today’s recovery processes.
You’re going to hear the word “agent” a few hundred times in Las Vegas. Almost all of it will be about building them. Very little will be about what happens when one does something it shouldn’t.
Agents hold credentials. They call APIs. They read and write production data at a speed no human can match, with permissions you gave them. A compromised agent, or just one that confidently does the wrong thing, can delete or exfiltrate or corrupt data faster than most ransomware.
We think about agent resilience the same way we think about any other workload, with a few extra wrinkles. You should snapshot what agents touch so you can roll back the moment behavior goes sideways. Watch for abnormal agent activity in the backup data, the same way you’d watch for encryption patterns. And make sure recovery doesn’t depend on the agent explaining what it did, because it may not know. Only 3% of organizations told us their recovery plans are ready for attacks that use frontier AI. The agents inside your environment deserve at least as much scrutiny as the attackers outside it.
A year ago at this same event, we announced that Cohesity would be a launch partner for the AWS European Sovereign Cloud. At the time, sovereignty was something European customers brought up near the end of a meeting. Now it comes up in the first five minutes, and not only in Europe.
The AWS European Sovereign Cloud is a dedicated environment built for data residency and operational independence and Cohesity Data Cloud runs inside it with the same controls we rely on everywhere else: immutability, multi-role access, quorum approvals, air-gapped vaulting.
The question regulators ask under DORA and NIS2 is the same one FedRAMP buyers ask in different words: where does your recovery copy live, who can reach it, and can you prove you can recover without crossing a border you’re not allowed to cross? If the honest answer is “we’re not sure,” that’s worth a conversation.
This is the one we care most about, because it’s where most recovery plans quietly fail. They answer the question “Can we restore our data?” They don’t answer, “Can we run the business while we do?”
Those are different questions. Our 2026 report found that 78% of organizations focus cyber recovery on restoring systems rather than keeping the business running. It’s easy to understand why. Restoring systems is measurable and it’s what the tools do. But a hospital with restored servers and no working identity provider still can’t admit a patient. A bank with its databases back and its payment gateway in quarantine still can’t settle a trade.
That’s why you define and operationalize your Minimum Viable Company. It’s the smallest set of services, people, systems, data, and decision-making you need to keep critical operations going after a destructive attack. Not everything. The things that matter most. Identity goes first, always, because nothing authenticates until Active Directory or Entra ID or Okta or Ping is back. It’s also the step teams are most likely to shortcut: 68% of organizations that were attacked in the past 12 months admit they don’t always test identity systems afterward to confirm the attacker’s persistence is gone. Then the MVC applications and their data and dependencies, verified clean. Then your security tooling so you can watch the rebuild. Then everything else, in whatever order the business decides.
The numbers here are humbling. Only 37% of organizations have written their MVC down. Only 22% have written it down and tested it. But in that 22%, 64% said their MVC determined what got restored first when a real attack hit. The plan worked for the people who had one.
An MVC you haven’t tested is a guess. An MVC without a vault behind it is a wish. You don’t need to vault everything. But you need to vault what is most essential for your MVC.
Find us at Booth #1874. It helps if you bring three numbers: your current RTO, the date of your last full recovery test, and how many agents you have in production. We’ll be honest about where the gaps are and show you what closing them looks like on AWS.
And if you take one idea home, make it this: start measuring time to restoring your Minimum Viable Company. Not time to restore a server. Time until the business can take an order again. Measure it every time you test. It’s the number your board wants, even if they don’t know how to ask for it yet.
Written By
Jared Ruckle
Sr. Director - Product Marketing, Solutions & Industry
Joyce Tse
Product Marketing Manager
Last updated: November 18, 2026