When AI enters public life, it rarely arrives as something a government built. It arrives as something a government bought.
The procurement contract, not the ethics charter, is where rights, transparency, and accountability are either written into a system or quietly lost. That makes public procurement one of the most overlooked levers in AI governance, and one of the most decisive.
The gap: principles without tools
Governments are the largest single purchasers of AI, yet they are buying it without the tools to buy it responsibly. The WSIS+20 outcome document, the Global Digital Compact, and the Council of Europe’s Framework Convention on Artificial Intelligence all advance principles. What is missing is the operational layer: the contract clauses, documentation requirements, audit rights, and risk assessments that decide whether those principles hold at the point of deployment. Principles are abundant. Practice is not.
From a UNESCO workshop to a CHI award
This work began in a room. At an AI & Equality by Women at the Table workshop at UNESCO in June 2025, the question was put directly: can public procurement be used to demand responsible AI? Over the following summer, researchers from the University of Cambridge and the Research Center Trustworthy Data Science and Security (RC Trust) turned that question into evidence.
The result, “It’s Just a Wild, Wild West”: Harnessing Public Procurement as an AI Governance Mechanism”, was accepted to CHI 2026, the premier conference in human-computer interaction, where it won an Honourable Mention. The findings then went back into practice, shaping procurement workshops with VNG, the association of Dutch municipalities, and other local-government bodies. A workshop question became peer-reviewed research and working tools in under a year.
What the research found
Formal, AI-specific procurement is rare in the EU and UK. Instead, AI slips into public services through channels with little scrutiny:
- Function creep: AI functionality added to existing systems through routine updates, often unnoticed by the buyer or the public.
- Hidden AI: AI embedded in physical products, from vehicles to traffic lights, collecting data without ever being declared as AI.
- Extended pilots: trials that run long and stay below the financial thresholds that would trigger oversight.
- Framework contracts: broad agreements with a few large, pre-selected vendors that streamline buying at the cost of fair competition and AI-specific scrutiny.
Four barriers keep the public sector from using its buying power: no clear guidance on how procurement law applies to AI; limited capacity to scrutinise vendor claims; the outsized influence of large vendors; and vendor lock-in, where AI layered onto legacy systems becomes almost impossible to replace.
A public trust framework for procurement
Against that, the research sets out concrete ways for public buyers to take back control, gathered into a public resource at compacctsys.net/procurement:
- A central, actionable AI vision, translated into operational guidance for procurement officers and standard contract clauses they can reuse.
- Knowledge sharing across government, through shared libraries and case studies, collaborative platforms, and joint procurement that pools bargaining power against large vendors.
- Interdisciplinary teams and upskilling, so legal, technical, and social-science expertise sit at the same table, and procurement shifts from administrative efficiency to strategic governance.
- Outcome-focused tenders that define the problem to be solved rather than prescribe the product, with early dialogue with affected communities.
- Data and IP governance and open-source reuse, keeping public control over data inputs and ownership of the resulting models.
- Lifecycle quality control: mandatory human rights impact assessment, ethical and social impact weighted in selection, and flexible contracts that can adapt as risks emerge.
Together these turn “trustworthy AI” from a slogan into the terms of a contract.
From research to standard
None of this stays on the page. At the WSIS+20 Forum on 6 July 2026, we convene “Public Trust and AI Procurement: From Principles to Practice” with the University of Cambridge, RC Trust, and the Council of Europe, putting procurement officers, legal teams, and policymakers in the same room to translate commitments into contract language.
The timing is deliberate. The Council of Europe’s Steering Committee for New and Emerging Digital Technologies (CDNET) is mandated to produce AI procurement guidelines by the end of 2026. Our framework and evidence are built to inform them, carrying a question first asked at a UNESCO workshop all the way to a binding-track standard.
That is Public Interest Tech in practice. Not waiting for principles to trickle down, but building the operational tools that decide whether public AI serves the public.