AI Procurement
Knowledge Bundle: Public-Sector AI Readiness Concept ID: government:ai_procurement Concept Type: Procurement Guidance Status: Draft Audience: Procurement staff, IT teams, legal counsel, department leaders, vendors Steward: OKF Expert AI Use: Human-reviewed Last Reviewed: June 2026
Purpose
AI procurement is the process of evaluating, purchasing, contracting for, and managing artificial intelligence tools or services in a way that protects the agency, the public, and the integrity of government operations.
AI tools should not be purchased solely because they are popular or impressive. They should be evaluated based on the problem they solve, the risks they introduce, the data they require, and the agency’s ability to govern their use.
Procurement Questions
- What specific business problem will the AI solution address?
- Is AI necessary, or would a simpler tool solve the problem?
- What data will the vendor collect, process, store, or retain?
- Will the vendor use agency data to train its models?
- Where will data be stored?
- What cybersecurity standards apply?
- What privacy protections are required?
- Can the agency audit or review the vendor’s performance?
- Can the agency export its data if the contract ends?
- What happens if the vendor changes pricing, features, models, or terms of service?
- What human review and accountability requirements apply?
Recommended Procurement Requirements
- Clear scope of work
- Data ownership provisions
- Data retention and deletion requirements
- Confidentiality and privacy protections
- Cybersecurity requirements
- Incident notification requirements
- Limits on vendor model training using agency data
- Accessibility requirements
- Audit and reporting rights
- Performance measures
- Exit and transition provisions
- Requirement for human oversight where appropriate
Vendor Evaluation Categories
| Category | Questions to Ask |
|---|---|
| Security | How does the vendor protect data, manage access, and respond to incidents? |
| Privacy | What data is collected, retained, shared, or used for model training? |
| Transparency | Can the vendor explain how the system works and what its limitations are? |
| Reliability | What happens when the system produces inaccurate, incomplete, or harmful output? |
| Accessibility | Does the tool support accessible public services and staff use? |
| Interoperability | Can the agency export data and integrate with existing systems? |
| Governance | Does the vendor support audit logs, administrative controls, and human oversight? |
Procurement Record
For each AI purchase, the agency should maintain a record including:
- Proposed use case
- Responsible department
- Vendor name
- Data involved
- Risk level
- Required approvals
- Contract protections
- Review date
- Renewal or termination conditions