Public procurement is the most overlooked lever in AI governance — the moment where rights, transparency, and accountability are either written into a system or quietly discarded.
Government AI isn't built. It's bought.
What this is
Public authorities are the tech industry’s largest customers. That purchasing power is a strategic lever — to shape the market toward the public interest and toward digital sovereignty, rather than passively consuming whatever vendors supply. This is the operational framework for using it.
A collaboration between AI & Equality by Women at the Table, the University of Cambridge, and the Research Center Trustworthy Data Science and Security (RC Trust), grounded in interviews with procurement professionals across the EU and UK and published in the CHI 2026 Honourable Mention paper “It’s Just a Wild, Wild West.”
How AI actually enters public services
Formal, AI-specific procurement is rare.
Systems mostly arrive through channels with reduced oversight:
Function creep
AI added to existing services through routine updates, often invisible to buyers and end users.
Hidden AI
capabilities embedded in physical products like vehicles or traffic lights, collecting data without being declared as AI.
Extended pilots
prolonged trials kept below the financial thresholds that would trigger monitoring.
Framework contracts
broad agreements with a handful of pre-selected large suppliers, streamlining purchase at the expense of competition and AI-specific scrutiny.
By the time a system touches a citizen’s life — determining benefits, flagging risk, allocating resources — the decisions that mattered most were already made, in contract terms nobody scrutinised.
Four barriers
Lack of clear guidance
— no consistent vision for how procurement law applies to AI’s specific characteristics.
Limited capacity
— procurement teams rarely have the technical footing to scrutinise vendor claims or spot long-term risk.
Strong vendor influence
— large firms exert outsized influence through polished pitches and existing relationships.
Vendor lock-in
— AI layered on legacy systems makes switching or decommissioning extremely costly.
The six practices
Six ways to shift the balance back to the public interest — each with the concrete mechanisms that put it into effect.
1. Clarity on Vision
A central, actionable vision that defines AI and its intended use, so the public sector can advocate for citizen values from the outset.
Mechanisms:
· a high-level strategic document setting the purpose and ethical boundaries of AI use · operational guidance translating that vision into specific tenders · pre-written standard clauses for transparency and accountability.
2. Knowledge Sharing
Pool expertise and demand across levels of government rather than negotiating alone against the same dominant vendors.
Mechanisms:
· hubs and libraries of procurement documents, case studies, and technical evaluations · collaborative platforms to share experience and coordinate requirements · joint procurement that pools demand to increase bargaining power.
3. Upskilling Culture
Prepare people for careful AI procurement by building interdisciplinary teams and AI literacy.
Mechanisms: interdisciplinary teams combining legal, technical, and social-science expertise · training to identify AI-specific risks and scrutinise vendor claims · a mindset shift from administrative efficiency to strategic governance and public value.
4. Outcome-Focused
Define the problem to be solved rather than prescribing the technical solution, leaving room for innovation.
Mechanisms: early stakeholder dialogues with affected communities to define the real problem · pre-tender market consultations on what’s technically possible · flexible procedures such as Innovation Partnerships for complex needs.
5. System Interactions
Treat AI as part of a wider infrastructure: strengthen data governance and consider open-source reuse.
Mechanisms: data and IP governance keeping public control over inputs and ownership of resulting models · prioritising modular, auditable, open-source components · mapping how new tools interact with legacy systems and dependencies.
6. Quality Control
Build responsible-AI requirements and monitoring across the lifecycle, not just at signing.
Mechanisms: tender criteria that weight ethical and social impact heavily · mandatory human rights and public-service impact assessments · flexible contracts that allow adjustment as systems evolve or risks emerge.
From research to standard
A question first asked at a UNESCO workshop became peer-reviewed research, became working tools, and now feeds the bodies setting standards:
→ practitioner interviews across the EU and UK
→ the CHI 2026 Honourable Mention paper
→ tested in workshops with municipalities including VNG, the association representing all Dutch municipalities
→ contributing to the Council of Europe’s CDNET working group on AI procurement guidance.
The WSIS vision of people-centred digital societies won’t be delivered by principles alone. It will be delivered — or betrayed — contract by contract, clause by clause.
Key terms
- Public procurement — how public authorities purchase products and services from companies.
- Digital sovereignty — a state’s ability to control its digital destiny: data, infrastructure, and the technology it depends on.
- Vendor lock-in — dependence on one vendor that can’t be undone without substantial cost.
- Function creep — the gradual widening of a technology’s use beyond its original purpose.
- Pilot projects — small, short-term experiments testing feasibility before full deployment.
- Framework contracts — broad supplier agreements setting terms for future purchases, used to streamline procurement.
The team
Anna Ida Hudig
University of Cambridge
Emma Kallina
Public Interest Tech Lead, AI & Equality by Women at the Table; RC Trust (UA Ruhr / University of Duisburg-Essen)
Jatinder (Jat) Singh
University of Cambridge; RC Trust
Caitlin Kraft-Buchman
Founder & CEO, Women at the Table / AI & Equality
Publication
Hudig, A. I., Kallina, E., & Singh, J. (2026). “It’s Just a Wild, Wild West”: Harnessing Public Procurement as an AI Governance Mechanism. Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (CHI ’26), Article 35, 1–22. doi.org/10.1145/3772318.3791968 — CHI 2026 Honourable Mention.