Causal Telemetry and Guarded AI Agents for Self Healing Object Storage

Self-healing storage is often described as an anomaly detector connected to an automation tool. That design is incomplete. An anomaly establishes that behavior changed; it does not establish why the change occurred, whether intervention is safer than waiting, or whether the proposed action will preserve durability and availability. This paper presents a reference architecture for self-healing object storage built on causal telemetry and guarded AI agents. The system constructs a time-aligned evidence graph from metrics, logs, traces, topology, and change events; evaluates competing hypotheses; and emits a typed remediation contract rather than an administrative command. A deterministic authority layer rechecks preconditions, policy, blast radius, and rollback before an idempotent executor acts. Verification telemetry then determines whether the intended outcome occurred and whether secondary harm emerged. The approach supports progressive autonomy while retaining explicit human oversight for high-impact changes. It is an architectural proposal, not a report of production deployment or measured performance.

Artificial Intelligence Cloud Computing Software Engineering