For AI Vigilance to function as an institutionalized Function, a Universal Evaluation Baseline is required to establish the minimum level of Evaluation applicable across AI Systems, regardless of their type or form of deployment. This Baseline is not limited to a particular Company, Model, Technical Approach, or Deployment Model, but provides a common Evaluation Foundation across a broad range of AI Systems, including Open Models.

However, a Universal Baseline does not mean imposing the same Vigilance Capacity or Evaluation burden on every AI System. AI Systems differ significantly in Risk, Domain, Operational Context, Capability, Scale, and Systemic Impact, and therefore require different levels of Evaluation depth and expertise. What is standardized is the minimum necessary Evaluation Function, upon which additional Advanced Evaluation is applied according to Risk.

For Specialized / High-Risk AI, Advanced Evaluation is selectively applied where additional expertise is required. Healthcare AI may require Medical Domain Expertise, Critical Infrastructure may require Cybersecurity and Operational Expertise, and Systems involving Children may require Child Safety expertise. Rather than maximizing Risk Expertise, Domain Expertise, Operational Expertise, and Regional & Cultural Expertise uniformly, the Baseline is maintained where sufficient, while areas requiring additional specialized Evaluation are strengthened.

For Frontier AI, greater Scale, Capability, Scope of Use, Cross-domain Interaction, and potential Systemic Impact may require more comprehensive Advanced Vigilance. Evaluation must extend beyond individual Risks or Domains to examine interactions across multiple areas, Regional / Cultural Differences, Operational Context, Emerging Capabilities, and other system-wide factors. In this case, Advanced Evaluation is structured not as a limited additional function, but as Comprehensive Advanced Vigilance.

Accordingly, this structure does not represent a uniform Approach in which every AI System is monitored at the same intensity.

Universal Baseline → Risk-based Advanced Evaluation → Comprehensive Advanced Vigilance

This layered structure establishes a common minimum Evaluation Foundation across AI Systems while adding the necessary Vigilance Capacity according to Risk, Domain, Operational Context, Capability, Scale, and Systemic Impact.

It avoids imposing excessive Evaluation burdens on lower-Risk AI while ensuring appropriate Expertise for Specialized / High-Risk AI and broader, deeper Evaluation Capacity for Frontier AI. What should be universal is not the intensity of Evaluation, but the existence of a Vigilance Baseline.