Artificial-intelligence procurement requires more than asking whether a supplier “uses artificial intelligence”. The buyer needs to understand where it is used, what it affects, which organisations and services depend on it, and which evidence will remain available during the contract.
A proportionate evidence request should cover
- intended purpose and excluded use;
- supplier, model and service dependencies;
- data sources and processing;
- user and affected-person groups;
- evaluation and acceptance criteria;
- human oversight and escalation;
- security, identity and access;
- transparency and user information;
- monitoring, incidents and corrective action;
- model, prompt, data and supplier change;
- subcontractors and exit;
- contract evidence and audit rights.
The depth should reflect the use case. A drafting assistant and a system influencing access to a public service do not require identical evidence.
Buyers should distinguish four states
- evidence supplied and verified;
- evidence supplied but not independently checked;
- planned control;
- unsupported assertion.
The procurement should also define what happens after award. Contract evidence should be refreshed when the system, model, supplier, intended purpose or material risk changes.
United Kingdom procurement guidance provides optional disclosure questions and encourages proportionate due diligence. European public buyers have model contractual clauses for high-risk and non-high-risk systems. Those sources support a structured approach, but they do not remove the need for legal and sector-specific review.
DDAI view: procurement should create the first lifecycle evidence record, not another questionnaire that becomes obsolete at signature.