As AI agents take on more business-critical work, organizations need a strategy for recovering when automated actions disrupt applications, data, or supporting services. Delegates will learn how operational resilience, agent-aware recovery context, and protected backup copies can work as independent but complementary layers to keep AI services productive and recovery options open. Anthony Dutra of HPE presented this session for HPE Storage at Security Field Day 16, with David Paquette joining for delegate questions.

Dutra argued that rogue agents rarely cause traditional crashes. Instead, they make fast, invisible changes while everything appears to run normally, so the blast radius is hard to identify and rollback becomes all or nothing. HPE’s answer pairs two layers. HPE Zerto Software provides continuous, VM-level protection with a recovery journal. HPE StoreOnce adds policy-based, deduplicated backup that is immutable, independent of Zerto’s 30-day journal limit, and compatible with Commvault, Cohesity, and Veeam. The newest piece is a recently released Zerto plugin that reads OpenClaw and NemoClaw logs through the Zerto API, not MCP. When an agent acts on a protected VM, the plugin tags the corresponding journal checkpoint with that agent’s name. Checkpoints arrive every five to ten seconds, so an administrator can roll a virtual protection group or individual files back to just before the agent’s change. Rollback is triggered manually from the dashboard or through a script against the API. An online delegate asked how HPE prevents poisoned data from being replicated. Paquette acknowledged that it does not. HPE instead relies on fine-grained checkpoints tagged by its own encryption detection, by agent activity, and by partner integrations such as CrowdStrike, Microsoft Defender, and SentinelOne.

Dutra’s recommended response to an agent incident is to stop and contain the agent, look for a suitable Zerto checkpoint, and fall back to StoreOnce copies when the damage predates the journal window. Tom Hollingsworth of The Futurum Group returned to the earlier post-quantum discussion and suggested a service that would ingest data at rest, re-encrypt it with PQC algorithms, and write it back in place like a scheduled job. Dutra said such a mechanism is likely still in the roadmap phase. Paquette added that Zerto supports recovery between most hypervisors and to AWS and Azure. HPE Morpheus’s HVM is currently limited to moving workloads to and from VMware, with broader support planned.

Presented by Anthony Dutra, Technical Marketing Engineer. Recorded live at Security Field Day 16 in San Jose, CA on October 29, 2026. Watch the entire presentation at https://techfieldday.com/appearance/hpe-storage-presents-at-security-field-day-16/ or visit https://techfieldday.com/event/xfd16/ or https://HPE.com/ for more information.