AI Vigilance is not limited to a particular AI Model or technical approach. It applies across the Operational Scope in which AI participates in Decisions or Actions, including Foundation Models / LLMs, AI Applications, Physical AI, and Agentic AI that autonomously combines multiple processes and Actions. However, the Observation Points required for Vigilance differ according to the characteristics of each System.

For Foundation Models / LLMs, the primary Observation Points include Model Behavior, Output, Model Change, and Response Patterns. Continuously observing Response tendencies to identical or similar Inputs, Behavioral changes following Model Updates, and recurring Outputs under particular conditions makes it possible to detect Shifts and Emerging Patterns that may not be visible from a single Response.

For AI Applications, Observation Points extend beyond the Model itself into the actual environment of use. Decisions, Interactions, User Context, and System Behavior are observed within the Operational Context, including the Application Purpose, connected Data, Users, Authority, and Workflow. Even where the same Foundation Model is used, these conditions may produce different Risks and Outcomes.

For Physical AI, Observation Points extend to Sensor Input, Control Decision, Physical Action, and Environment / Outcome. Where AI Output does not remain informational but affects the Physical World through Machines or Robots, Vigilance must observe not only the Decision, but also what Action was actually executed and what occurred in the Environment as a result.

For Agentic AI, Observation Points may span multiple System Layers and Action Stages. Planning, Tool Use, Authority, and Multi-step Behavior must be traced to observe not only individual processes, but also how a sequence of Actions was formed and under what Authority it was executed. Agentic AI should therefore be understood not as requiring a separate type of Vigilance, but as a configuration in which multiple Observation Points are monitored across the system.

Importantly, although Observation Points differ according to the System being observed, the fundamental Functional Structure of AI Vigilance remains unchanged. The common structure of Observe / Monitor, Detect, Evaluate, Respond, Analyze, and Report is maintained, while appropriate Observation Points are defined according to System Architecture, Risk, and Operational Context.

This makes it possible to address Foundation Models, Applications, Physical AI, and Agentic AI under a common AI Vigilance Framework, rather than constructing separate Monitoring Frameworks for each technical approach.