Overview
Public-sector AI procurement is a lifecycle problem, not a completed award: agencies must define mission-specific requirements, secure usable data and intellectual-property rights, choose routes by the control they preserve, and test the system in its real workflow. Monitoring, human review, transparency, exit, and documented lessons keep changing services governable. The evidence supports adaptive, proportionate controls, not a universal testing standard, monitoring schedule, or validated framework.
Learning outcomes
- A contract award creates leverage only when requirements, evidence, rights, and responsibilities are usable before deployment.
- Testing must continue in the agency’s real workflow as models, data, users, and operating conditions change.
- Speed and flexibility are valuable only when the acquisition preserves proportionate testing, transparency, portability, cost visibility, and exit.
- Institutional learning turns incidents, limitations, and retirements into context-rich evidence rather than universal rules.
Keywords
Preview every chapter · Unlock full text, audio, and video from chapter 4
PlansFAQ
Why is a public-sector AI contract award not the end of procurement?
Because an award establishes a supplier relationship but does not prove that a changing capability will remain fit for the mission or that the agency can measure, challenge, change, or stop it.
How should agencies choose an AI acquisition route?
Agencies should choose the route by mission specificity, capability volatility, consequence and reversibility of error, and the control terms the route can carry—not by speed or flexibility alone. Different routes may fit different conditions, with compensating safeguards where needed.
What makes AI testing useful after deployment?
Post-deployment testing is useful when it compares current performance with an agency-defined baseline using mission-relevant data and conditions, while examining errors, overrides, cost, security, and public outcomes as appropriate.
What does adaptive governance require from a public agency?
Adaptive governance requires a proportionate way to detect material changes, assign responsibility, intervene, preserve records, and learn from evidence. It combines conditional controls, defense in depth, sociotechnical implementation, and revision without claiming a universal standard.