Toolkit · Instrument 3
Procurement Checklist
Most municipal AI arrives through vendor products, contract renewals, software updates, pilots, and procurement processes rather than internal development. That makes procurement one of the most important AI governance checkpoints a city controls.
Vendor disclosure questions
Adapt these questions for RFPs, pilots, contract renewals, amendments, and purchases of software with AI-enabled features. Require written responses before approval. If a vendor cannot provide clear answers, that is itself relevant information for the city's review.
- 01
Does this product use AI, machine learning, generative AI, automated prediction, ranking, or decision-making in any feature? List each feature.
- 02
What data does the system collect, receive, generate, infer, or retain? Is any city or resident data used to train, fine-tune, evaluate, or improve vendor models?
- 03
Can the vendor or a third-party model provider change system behavior after contract signing? What advance notice, documentation, testing, and city approval are provided?
- 04
What are the known error rates, limitations, performance boundaries, and failure modes of each AI-enabled feature?
- 05
Has the system been evaluated for performance differences or disparate impacts across relevant demographic or affected groups? Provide available methodology and results.
- 06
What logs, documentation, version information, testing access, audit rights, or explanations can the city obtain for consequential outputs?
- 07
Where is data processed and stored? Which subcontractors, cloud providers, model developers, or third-party services are involved?
- 08
What happens to city data, derived data, prompts, outputs, embeddings, logs, and system configurations at contract termination?
Apply the full checklist to Tier 3 and Tier 4 systems. For lower-risk tools, prioritize the questions most relevant to data protection, security, public records, system changes, performance, and vendor accountability.
Core contract clauses
| Clause | Purpose |
|---|---|
| Data reuse restriction | The vendor may not use city or resident data to train, fine-tune, evaluate, or improve models for unrelated purposes or other customers without written city authorization. |
| Change notification | The vendor must provide advance notice of material changes to AI features, model providers, system behavior, data practices, or performance, and allow the city to re-review the system. |
| Audit and logging | The city receives logs and documentation sufficient to review relevant inputs, system version, settings, human actions, and operational history associated with consequential outputs. |
| Records compliance | The vendor must support the city's applicable records-retention, disclosure, litigation-hold, and public-records obligations. |
| Incident reporting | The vendor must report material security incidents, harmful outputs, significant performance failures, or materially incorrect results within a defined period. |
| Performance and acceptance testing | The city may test the system before full deployment and reject, suspend, or require remediation if agreed performance, accessibility, security, accuracy, or reliability standards are not met. |
| Suspension and termination rights | The city may suspend or terminate use when the system causes material harm, fails required testing, changes substantially, creates unaddressed legal or security risk, or no longer complies with city policy. |
| Exit and portability | City data must be returned or securely destroyed at termination, with written certification and reasonable support for migration to another system. |
Model language should be reviewed and adapted by the city attorney and relevant procurement, records, privacy, security, and operational staff. These provisions are starting points, not legal advice.
Next step
AI Use and Transparency Policy
Once procurement safeguards are established, define how resident-facing AI systems will be disclosed, documented, reviewed, and contested.