Public Sector Rules and Transparency Duties When Buying AI
If you work in the public sector, buying AI systems comes with additional obligations beyond those faced by private organisations. These obligations arise from public law requirements for transparency, accountability and fairness in government decisions.
The EU AI Act Article 4 requires AI literacy training for staff using AI systems. Public sector staff must understand what AI can and cannot do. Your organisation must maintain records that this training occurred. This means: identify who uses AI systems, provide mandatory training (which could be this course or similar), and document that training was completed.
The EU AI Act Article 50 requires transparency when AI systems make decisions affecting individuals. If a public authority uses an AI system to decide whether to grant a benefit, license or permission, the affected individual must be told that AI was involved and must have the right to explanation. A person denied a business licence can say: “Tell me why. What information did the AI consider? What is the basis for the decision?” You must provide a comprehensible explanation.
This transparency requirement affects procurement. When evaluating AI systems, you must ask vendors whether the system produces explanations that are human-understandable. The system must provide specific reasons for each decision, not just performance statistics. If a vendor cannot provide such explanations, you cannot use their system for public decisions under Article 50.
A practical example: your local authority uses an AI system to assess social housing applications. The system scores applications and makes recommendations. An applicant is rejected. Under Article 50, you must tell the applicant why. “The AI system scored your application 45 out of 100” is not sufficient explanation. You must say: “Your application scored lower because you reported income below the median for your region, and you have no dependents. The system weighted income and family size as primary factors in assessment.” This explains the decision in comprehensible terms.
The EU AI Act also specifies that high-risk AI systems (embedded AI that affects rights) used by public authorities must be approved before deployment. Some system types are presumed high-risk for public sector use. Automated decision-making systems that deny benefits are examples. You cannot just buy a system and deploy it. You must follow a formal approval process, document your risk assessment and maintain records for regulators.
The UK also has transparency duties for public bodies. If a public sector organisation makes a significant decision using AI, the decision must be explainable and defensible. A tribunal or court reviewing the decision must be able to understand how AI contributed to it. This means procurement must prioritise explainability.
Public sector procurement also has value-for-money requirements. You cannot buy the first system that works; you must demonstrate that you chose the option that provides best value. This means you must compare vendors on cost, capability, risk and total cost of ownership, not just initial price.
A second consideration is data. Public sector organisations often hold sensitive citizen data. When buying AI systems, you must ensure the vendor has appropriate security and data handling practices. You should ask: “Will you store our data in the UK or EU? How is it encrypted? Can you provide security audit reports? Have you experienced breaches? What is your incident response process?”
Public procurement rules also apply. In the UK and EU, public bodies must follow procurement rules when buying goods or services. For large purchases (above a threshold around £200,000), you must go through a formal tender process and invite bids from multiple vendors. You cannot simply select the vendor you prefer. Tender rules ensure competition and transparency.
This affects AI procurement because it slows decision-making. A private company might select a vendor in weeks. A public body must go through a tender process that might take four to six months. Your project planning must account for this timeline.
Public sector procurement rules also impose requirements on vendors. They must meet security standards, prove financial viability and submit to audits. Some private vendors specialising in consumer markets are uncomfortable with public sector procurement rules because they require more documentation and accountability than consumer vendors are used to.
Practical example: a central government department wants to buy an AI system for benefits administration. The system will make decisions affecting eligibility for payments. The department cannot simply hire a vendor. It must: assess the AI system’s impact on beneficiaries, formally approve the system as high-risk, ensure staff are trained, ensure the system produces explainable decisions, follow public procurement rules (likely a formal tender), document compliance with data protection law and maintain an audit trail of decisions for regulators.
This process is longer and more expensive than private sector procurement, but it is necessary to ensure public accountability.