Subverting AIOps Systems Through Poisoned Input Data

This summary describes a 2025 academic paper showing that LLM-driven AIOps systems can be subverted via poisoned telemetry data: researchers present an automated attack pipeline (AIOpsDoom) that injects adversarial telemetry to induce harmful remedial actions, and propose a defense (AIOpsShield) to sanitize telemetry and mitigate such attacks—highlighting AIOps as an emerging attack surface with potential for infrastructure compromise.