AI misalignment is a failure mode where a system’s behavior diverges from the operator’s intended goal, rules, or safety constraints. Instead of doing exactly what was asked, the model may optimize the task in an unexpected way, overreach its permissions, or follow instructions that conflict with policy.
In cyber security, misalignment matters because agentic AI can browse websites, process untrusted content, and trigger tools or workflows. If its guardrails are weak, an attacker may steer it with crafted pages, prompts, or data, causing unintended actions such as exposing information, changing settings, or taking repeated actions without proper approval. Defenders reduce this risk with least privilege, tight tool permissions, human confirmation for high-risk steps, strong logging, and clear separation between trusted instructions and untrusted input.