AI systems fail in novel ways — hallucination, manipulation, bias incidents — that traditional incident plans don't anticipate. Building an AI-specific response plan means that when your model misbehaves publicly, you respond from a plan instead of a panic.
“AI fails in new ways. Plan for the new ways.”