AI Agent Safety Needs Boundaries and Trusted, Recoverable Data
Veeam, Tuesday, October 27th, 2026
Veeam argues AI agent safety requires enforced boundaries plus trusted, recoverable data, citing its 2026 resilience report.
Veeam's Rehan Jalil argues AI agent safety needs both enforced operational boundaries and a trusted, recoverable data layer.
He welcomes NVIDIA's Open Agent Safety Platform, which sandboxes and monitors agents, noting Veeam is in its launch ecosystem, but says agents acting within permissions can still overwrite or delete critical data.
Citing Veeam's Data Trust and Resilience Report 2026, 88% of organizations use or pilot AI agents, but only 40% of leaders are very confident they could isolate and reverse an agentic AI failure.
He recommends assigning an owner for agentic AI risk, inventorying agents and their data access, setting and testing AI recovery objectives, and benchmarking readiness.