Physical AI Autonomy refers to a dependability layer built around vision-language-action and world models, rather than a capability those models must provide alone. It maintains updated knowledge of an uncertain environment, monitors plans and actions as they execute, and detects when behavior diverges from expectations. When conditions depart from training, it adapts execution, enforces operational constraints, and incorporates human guidance rather than failing silently. The goal is autonomy trusted to operate around people in environments that are variable and only partly predictable.