Before deploying any high-impact use case the agency completes and documents an AI impact assessment, updated periodically and after significant modifications, covering the intended purpose and expected benefit with metrics or qualitative analysis against existing processes; the quality and appropriateness of the data and model capability with a summary of the data used across design, development, training, testing and operation, its collection and preparation, whether it will be released as open data and any information about federally protected classes; the potential effects on privacy, on civil rights and on civil liberties, and the effects of using or forgoing the AI, with planned mitigations and references to privacy impact assessments and waivers; reassessment schedules and triggers; a cost analysis of direct costs and expected savings; the results of an independent review by an agency reviewer not involved in the development, whose comments go to the person accepting the risk; and risk acceptance signed by that person.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.