For six days, something with valid credentials has been rewriting rows in your production database. It could be an attacker encrypting one column at a time, or an AI coding agent that misread a cleanup task, and the app kept serving traffic the whole time.
Your backups did their job. Every nightly copy sits in an immutable vault where nobody can touch it, and every one of them faithfully preserved the damage.
So the hard part of ransomware recovery is knowing which copy is clean, and when the data stopped being clean. That’s the question that separates enterprise backup solutions in 2026, and the gap between them is wide.
Here’s where each one lands:
Eon is the best enterprise backup solution for ransomware recovery in cloud estates, since it reads the logical contents of database backups and correlates multiple independent signals into one weighted score (add-on).
Database detection is still in early access, but the workflow is the one you want: it points to the last clean recovery point and rolls back only the records that were hit, including damage from a rogue AI agent.
Rubrik brings threat hunting from the security side, Commvault adds cleanroom rehearsals, and Dell PowerProtect centers on isolated Cyber Recovery vaults. Cohesity serves large data-center estates, Veeam is built around VM recovery, and Druva runs detection as a managed service.
| Solution | What its detection reads | Isolated recovery option | Licensing note |
| Eon | Database rows, VM files, object access (add-on) | Logically air-gapped vault | Usage-based; detection priced separately |
| Rubrik | Snapshot anomalies, IOCs, file hashes | Recovery Simulation | Hunting in Enterprise Edition |
| Commvault | File and VM signatures, encryption | Commvault Cleanroom | Threat Scan metered per FETB |
| Dell PowerProtect | Full content of vaulted copies | PowerProtect Cyber Recovery vault | By quote |
| Cohesity | Snapshot anomalies, threat feeds | FortKnox vault | Hunting in Enterprise; FortKnox add-on |
| Veeam | Block metadata, file activity | Veeam Data Cloud Vault | YARA from Advanced edition |
| Druva | Ransom notes, extensions, entropy | Copies outside your AWS org | Add-on on Enterprise and Elite |
Why isn’t an immutable backup enough?
Immutability guarantees that nobody can change or delete a backup copy. It says nothing about what’s inside, so an immutable copy of an encrypted or corrupted database is still encrypted or corrupted, just permanently.
That moves the 2026 question from whether a copy survived to whether it’s clean, and when it stopped being clean. Answering it takes detection that reads the right layer of data, plus a way to restore only what changed so legitimate writes made after the attack survive.
Managed databases make this harder. Services like Amazon RDS and Aurora never hand you their underlying storage files, so detection for them has to read the logical contents (rows, tables, and schemas) through the database itself.
1. Eon: best for clean-point recovery of cloud databases
Pick it if: your estate runs on AWS, Azure, or Google Cloud, and the data you’d lose sleep over lives in RDS, Aurora, Cloud SQL, or MongoDB Atlas. Built by the team behind CloudEndure, the platform handles cloud data protection and recovery.
Setup is a single read-only IAM role with no agents, appliances, or clusters, and the platform keeps discovering and classifying resources so policy applies without manual tagging.
Detection and clean-point recovery: Detection is a paid add-on that correlates multiple independent signals (such as row-count drops, schema changes, and ransom notes) into one weighted score.
Database detection, still in early access, reads the logical contents of a backup because managed services don’t expose backup files to scan. For VMs it weighs file entropy and malware signatures, and for object storage it flags mass deletion and abnormal access.
When something trips, the platform points to the last clean recovery point and restores only what changed: a table, the rows a SQL query matches, a file, or a MongoDB Atlas document. Later good writes stay put, so rolling back the exact records hit takes minutes.
It also spots AI apps and agents running on VMs and rolls back what a rogue agent changed. Recovery copies sit in an immutable, logically air-gapped vault with time-bound access, and restores need multi-party approval.
Licensing: It’s usage-based, billed per GB per month on protected storage, with no tier ladder and flexible spending commitments, and ransomware protection is a package on top. There are no feature gates, and global deduplication helps it reach up to 50% lower cost.
Trade-off: There’s no packaged cleanroom, so you pair it with an isolated environment you run. On-prem support is limited, so an estate that’s still mostly on-prem needs a separate tool for the data center, and the reference list is shorter than the incumbents can show.
2. Rubrik: strongest for security-led threat hunting
Pick it if: your security team wants to hunt threats inside backup history and treats recovery as part of incident response. Rubrik calls itself “the Security and AI Operations Company,” and the product follows suit.
Detection and clean-point recovery: Anomaly Detection monitors for encryption, unusual access patterns, and other ransomware indicators. Threat Monitoring scans backups for indicators of compromise using intelligence from Mandiant and Rubrik Zero Labs.
Threat Hunting matches file patterns, hashes, and YARA rules and analyzes a time series of snapshots to pinpoint a potentially clean one, a strong fit wherever scanning can read storage files, so ask how it applies to managed databases like Aurora or RDS before you sign.
Threat Containment can isolate infected snapshots, and Recovery Simulation tests plans in isolated environments. For AI mistakes, Agent Rewind is designed to undo what changed across files, databases, configurations, or repositories.
Licensing: As of October 2026 there’s no public price list, and every edition shows Contact Sales. Anomaly Detection starts in Business Edition, while Threat Monitoring, Turbo Threat Hunting, and Threat Containment sit in Enterprise Edition, sold as a 3-year subscription.
Trade-off: In the customer-hosted cloud model, you own the networking, IAM roles, and security posture. Rubrik Managed hosts the compute and moves data outside your AWS organization by design, which Rubrik says won’t suit a policy that keeps backups inside your own account boundary.
For RDS, Rubrik documents recovery at the instance level, to an exact point in time, so rolling back a few damaged tables means restoring the instance and extracting them.
3. Commvault: best for rehearsing recovery in a cleanroom
Pick it if: you run SAP or Oracle at scale and want a safe place to rehearse a cyber event before it’s real. Commvault’s SAP recovery goes down to a tablespace or a single data file.
Detection and clean-point recovery: Threat Scan layers Google Threat Intelligence, hash-based hunting, malware signatures, heuristics, YARA, and AI-enabled encryption detection to find and quarantine encrypted or corrupted files and VMs.
Its AI-assisted Synthetic Recovery assembles the most current verified clean files into a recovery-ready point, pitched as a way to surgically remove compromised data while recovering the rest.
Threat Scan’s documentation covers VM and file workloads, and Clumio, which Commvault acquired in 2024, prices ThreatDetect for EC2 and EBS. Both read files and volumes, the layer this style of scanning was built for.
Commvault Cleanroom then spins up isolated recovery environments on demand for testing, staging, and forensic analysis, in the cloud or on-prem.
Licensing: As of October 2026, Threat Scan is licensed and metered on front-end terabytes, and self-managed Commvault Cloud is quoted on the same unit. Commvault’s SaaS pricing page lists database backup starting at $90 per TB per month, and Clumio ThreatDetect Standard is $0.014 per GiB-month.
Trade-off: On customer-managed storage, Threat Scan needs access nodes you deploy and keep apart from backup proxies, and it analyzes files up to 50 MB. Commvault’s cloud deployment docs for Cleanroom name Azure and AWS, so a Google Cloud-centric team should confirm its recovery target early.
4. Dell PowerProtect: strongest for isolated vaults on Dell storage
Pick it if: you already run Data Domain and Dell primary storage like PowerStore, PowerMax, or PowerScale. PowerProtect One, launched in May 2026, builds on Data Manager and Data Domain, so cyber recovery extends gear you already own.
Detection and clean-point recovery: PowerProtect Cyber Recovery isolates critical data in a separately governed vault with its own controls, credentials, and management plane, either on-premises or in a cloud vault on AWS, Azure, or Google Cloud.
Dell Cyber Detect scans those vaulted copies with full-content analytics across more than 200 content-based statistics, and Dell says it was trained on 7,500+ ransomware variants.
Its forensic reports show blast radius, corrupted files, owners, and last known good backups, so your team picks a trusted recovery point before restoring anything.
Scope still tracks Dell hardware. Cyber Detect for Storage runs on PowerStore today, with PowerMax still listed as planned.
Licensing: As of October 2026, Data Manager and Cyber Recovery are sold by quote through Contact Sales, and the Data Domain family page lists no price. Compliance-mode Retention Lock carries its own license key.
Trade-off: The core is appliance-centric, with Data Domain systems from the DD3410 to the DD9910F that you deploy and run, and the deepest integrations assume Dell storage. For a cloud-first team, that’s a lot of infrastructure to stand up for one vault.
5. Cohesity: a fit for deep data-center footprints
Pick it if: you run a large data-center estate, possibly on NetBackup, and want cyber recovery without a forklift migration. Since the Veritas combination closed in December 2024, Cohesity tells NetBackup customers to “keep, extend, or transform” those environments.
Detection and clean-point recovery: Anomaly detection continuously monitors backup snapshots for abnormal data changes. Threat scanning layers Google Threat Intelligence and Sophos feeds, YARA rules, SHA-256 hash hunting, and a Google Cloud-managed malware sandbox.
RecoveryAgent scans candidate snapshots against threat intelligence, hash feeds, and historical anomaly data, then recommends a clean point with evidence behind it. NetBackup shows which backups were the last known good, malware-free copies.
FortKnox is SaaS on AWS, Azure, and GCP or self-managed on your own infrastructure, and a Minimum Viable Recovery Environment gives you a clean room to investigate in. Agent Resilience, announced in September 2026 for select customers, covers agent memory and the databases agents interact with.
Licensing: As of October 2026, Cohesity publishes no dollar pricing, and Core and Enterprise editions come as 1-, 3-, or 5-year subscriptions. Core includes Ransomware Anomaly Detection, Enterprise adds Threat Hunting and Cyber Recovery Orchestration, and FortKnox is an add-on.
Trade-off: DataProtect and NetBackup remain separate stacks under one vendor, and NetBackup keeps its own appliances, so a mixed estate ends up running both. Cloud Rebuild covers Terraform-provisioned environments and select AWS workloads only.
6. Veeam: best for hypervisor-heavy estates
Pick it if: most of what you’d recover after an attack is virtual machines. Veeam Data Platform v13.1 took support to 14 hypervisors in July 2026, and the Veeam Universal License moves between virtual, physical, and cloud workloads.
Detection and clean-point recovery: Inline scan analyzes the metadata of data blocks during backup, flagging encrypted data, file types, onion addresses, and ransomware notes. File system activity analysis watches for mass deletions and extension changes, with indicators drawn from MITRE ATT&CK.
Threat Hunter adds signature scanning alongside YARA rules, and Scan Backup exists to find the last clean restore point when the attack date is unknown. Secure Restore scans data during the restore itself.
For AI agents, Veeam Intelligent ResOps for Microsoft 365 went GA on October 15, 2026, as an add-on that shows what an agent changed, file by file, and undoes it.
Licensing: As of October 2026, Veeam Data Platform is metered per workload with no published dollar figure. AI-powered malware detection is in all three editions, while YARA and Threat Hunter start at Advanced and orchestrated recovery is Premium.
Veeam Data Cloud Vault, which is always immutable and logically air-gapped, costs $4.50, $14, or $24 per TB per month across its Archive, Foundation, and Advanced tiers.
Trade-off: You run the backup server, appliances, repositories, and worker instances yourself. Scan Backup skips backups in a scale-out repository’s archive tier and storage snapshots, and the indicators-of-compromise list can’t be customized.
7. Druva: a fit for teams that want SaaS-delivered protection
Pick it if: you want protection across data center, cloud, endpoints, and SaaS apps with no infrastructure to run. Druva delivers it as a fully managed service and keeps backups in its own cloud, outside your AWS organization.
Detection and clean-point recovery: Ransomware Detection, in limited availability since September 2026 for EC2, EBS, VMware, and Azure VMs, looks for ransom notes, suspicious extensions, and mass file renaming, then checks hits with entropy and MIME type analysis.
Threat Watch scans changed snapshots multiple times a day and rescans the last 30 days when new threats emerge. Curated Recovery builds a golden snapshot from the latest clean file versions in the attack window.
Threat Hunting permanently excludes database file types such as db, mdf, and sqlite, which fits how file-level hunting works. Druva’s AI-agent rollback for Claude Code, launched in July 2026, covers locally stored data on endpoints.
Licensing: As of October 2026, Public Cloud plans are sold per TB per month after deduplication, often through credits where one credit equals one TB-month. Data Anomaly Detection, Quarantine Bay, curated recovery, and Threat Insights are paid add-ons on Enterprise and Elite.
Trade-off: The copy lives in a vendor-run environment, which buys isolation and gives up your own account boundary. In November 2026, Druva’s plans page still listed curated cyber recovery for Public Cloud workloads as coming soon.
Run the rollback drill before an attacker forces one
Pick one production database and schedule a timed rollback drill. Bring the on-call engineers who’d get paged at 2 in the morning, hand them a corruption window, and have them find the last clean point and restore only the affected tables.
Whatever slows them down, whether it’s a missing permission, an approver nobody can reach, or a restore that insists on rebuilding the whole instance, is your real recovery time. Fix that first, then run the drill again next quarter.
FAQs
How do you find the last clean recovery point after ransomware?
Work backward through restore points with detection that reads what the workload stores, such as file contents for VMs or row counts for databases, until signals look normal. The newest point before the anomaly is your candidate, and it still needs a test restore.
Can an AI coding agent cause a data-loss incident like ransomware?
Yes. An agent with valid credentials can delete or rewrite production data quickly, and because its actions look authorized, every new backup copies the damage forward.
Should ransomware recovery restore everything or only what changed?
Restore only what changed whenever you can identify it, because a full rollback discards every legitimate write made after the clean point. Full restores still make sense when the attacker compromised the system itself.
Why can’t file-based scanning check a managed database backup?
Managed services like Amazon RDS and Aurora don’t expose their underlying storage files, so a scanner that opens files has nothing to inspect. Detection for them has to read the logical contents, such as rows, tables, and schemas, through the database engine.