CoChat
Sign Up
Cited by 3Public Procurement and PolicyE-Government and Public ServicesEthics and Social Impacts of AI

“It’s Just a Wild, Wild West”: Harnessing Public Procurement as an AI Governance Mechanism

Open access
Hudig, Anna Ida, Kallina, Emma, Singh, Jatinder·2026·Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems
DOI 10.1145/3772318.3791968
Open this paper in CoChat — free account

Abstract

Public sector AI has the potential to harm citizens, with risks increasing as its use expands. Recent work positions public procurement as a way to shape public sector AI in line with public interests, using the state’s purchasing power to influence which AI systems are procured and under what conditions. This paper examines how this potential can be realised in practice by drawing on semi-structured interviews with UK and EU buyers, providers, and procurement experts. Our findings result in six promising procurement practices that enable the public sector to shape AI in line with public interests, alongside concrete mechanisms to support their uptake. Further, we find that AI-specific procurement approaches remain immature and systems often enter through informal channels with less scrutiny. We provide directions for both research and practice on how public procurement can be used as a governance mechanism for better aligning AI with public interests.

AI snapshot

Written from the paper’s full text.

TL;DR

The paper examines public procurement as an AI governance mechanism in the UK and EU through 16 expert interviews. It finds AI-specific procurement remains immature and AI often enters the public sector through informal channels with less scrutiny, while identifying six promising procurement practices and concrete mechanisms to better align public sector AI with public interests.

Key findings

  • Formal, AI-specific procurement is rare and in an early stage; AI is often folded into standard IT procurement, framework contracts, or informal routes such as function creep, hidden AI in connected products, and pilots structured below financial thresholds, affording less scrutiny.
  • The paper identifies six promising procurement practices: provide clarity on vision, guidance, and operationalisation; harness knowledge sharing and collaboration; prepare people and culture; focus on defining the outcome/problem rather than prescribing solutions; consider system interactions and reuse; and implement quality control measures including responsible AI requirements across the procurement lifecycle.
  • It foregrounds four core challenges: lack of clear guidance and weak embedding of public values and regulations; limited organisational capacity and AI procurement literacy; strong vendor influence and skewed competition; and complex infrastructures creating vendor lock-in.
  • The authors argue public procurement is an underutilised lever and outline an HCI research agenda, positioning procurement as a topic for HCI scholarship through participatory methods, AI literacy, socio-technical audits, and collaborative policy design.

Methods

  • Semi-structured interviews with 16 participants across the UK and EU, including public sector buyers, technology suppliers, lawyers, procurement researchers, consultants, and guidance authors; recruitment used convenience and snowball sampling, and participants had professional expertise in digital products and services procurement.
  • Interviews lasted 46–68 minutes, were conducted via Microsoft Teams, recorded with automatic transcription, and guided by an interview protocol covering current AI acquisition, challenges, and promising practices.
  • Data were de-identified (P1–P16) and analysed using the six phases of thematic analysis; the two lead researchers initially coded independently, compared codes and sub-themes, resolved conflicts to consensus, and iteratively analysed remaining interviews with ongoing discussion.

Limitations

  • The paper states the study is qualitative and exploratory among a relatively niche, hard-to-reach expert group, so findings are not statistically generalisable; the sample was slightly skewed towards the UK and the Netherlands but also included Italy, Germany, Belgium, Slovenia, and EU-level actors.
  • The paper states both lead researchers had prior professional interactions with the public sector, which facilitated access and contextual knowledge but risked bias; this was mitigated through reflexive discussions, independent coding, and collaborative analysis.
  • The text makes evident that convenience and snowball sampling may introduce self-selection or network bias, and the findings rely on participant reports rather than direct observation of procurement outcomes.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1145/3772318.3791968

Cites

139 of 139 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

29 of 64 cited sources checked · 9 supported · 0 flagged · accuracy not yet measured

  • Premisecitation
    § 2.3 Public procurement for AI governance: aspirations vs practice · p. 4

    In the UAE, procurement plays a critical role in AI adoption when success factors such as clear needs assessment, legal compliance, technical expertise, and supplier transparency are in place.

    “In the UAE, procurement has been shown to play a critical role in AI adoption when success factors such as clear needs assessment, legal compliance, technical expertise, and supplier transparency are in place [7].”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [7] [7] Khalid Alshehhi, Ali Cheaitou, and Hamad Rashid. 2024. Procurement of Artificial Intelligence Systems in UAE Public Sectors: An Interpretive Structural Modeling of Critical Success Factors. Sustainability 16, 17 (Jan. 2024), 7724. doi:10.3390/su16177724 Publisher: Multidisciplinary Digital Publishing Institute. - source

      Section 2.3 identifies critical success factors for successful AI procurement in the UAE public sector, and the paper lists clear needs assessment, regulatory compliance, and vendor technical capabilities among relevant considerations. It discusses transparency in procurement and AI systems, but does not identify supplier transparency as a success factor. The claim also extends the paper’s public-sector scope to AI adoption in the UAE generally. (LLM, conf=0.84)

      Experimental — accuracy not yet measured
  • Premisecitation
    § 1 Introduction · p. 1

    AI systems used in the public sector are often purchased from or outsourced to the private sector.

    “Often, such systems are purchased from private sector organisations [9, 94].”
    • Also stated in: § 2.2 Public sector AI: adoption, suppliers, and safeguards · p. 4 · plain
      “Importantly, the AI systems used in the public sector are often purchased from or outsourced to the private sector [9, 94].”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites [9] [9] Local Government Association. 2025. Responsibly Buying AI - How to build equality & data protection into your AI commissioning and procurement processes: a guide for councils in England . Technical Report. Local Government Association. -

      Not checked in the free run.

    • cites [94] [94] Nari Johnson, Elise Silva, Harrison Leon, Motahhare Eslami, Beth Schwanke, Ravit Dotan, and Hoda Heidari. 2025. Legacy Procurement Practices Shape How U.S. Cities Govern AI: Understanding Government Employees' Practices, Challenges, and Needs. In Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency (Athens, Greece) (FAccT '25) . Association for Computing Machinery, New York, NY, USA, 772-789. doi:10.1145/3715275. 3732049 - source
  • Premisecitation
    § [Emma Kallina ∗](https://orcid.org/0000-0003-4912-7216) · p. 1

    In the US, the private sector currently exerts disproportionate influence over the AI systems purchased by the public sector, including their capabilities and technical specifications.

    “A recent study in the US context finds that the private sector currently exerts disproportionate influence over the AI systems that are purchased by the public sector, including their capabilities and technical specifications [94].”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [94] [94] Nari Johnson, Elise Silva, Harrison Leon, Motahhare Eslami, Beth Schwanke, Ravit Dotan, and Hoda Heidari. 2025. Legacy Procurement Practices Shape How U.S. Cities Govern AI: Understanding Government Employees' Practices, Challenges, and Needs. In Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency (Athens, Greece) (FAccT '25) . Association for Computing Machinery, New York, NY, USA, 772-789. doi:10.1145/3715275. 3732049 - source

      The Introduction says most AI systems used by U.S. governments are developed and acquired from private vendors. Sections 5.1 and 5.2 report that vendors shape cities’ understanding of AI use cases, withhold technical information, and sometimes introduce features cities struggle to disable. These findings support substantial vendor influence, but the paper does not establish disproportionate influence over technical specifications across the U.S. public sector; its empirical findings concern interviewed cities. (LLM, conf=0.84)

      Experimental — accuracy not yet measured
  • Premisecitation
    § [Emma Kallina ∗](https://orcid.org/0000-0003-4912-7216) · p. 1

    Public procurement could position the public sector as a gatekeeper, ensuring AI systems purchased by government agencies are better aligned with the public interest.

    “According to these works, public procurement could position the public sector as a gatekeeper, ensuring that AI systems purchased by government agencies (where such systems typically have direct societal impact) are better aligned with the public interest [13, 94].”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites [13] [13] Dan Bateyko and Karen Levy. 2025. One Bad NOFO? AI Governance in Federal Grantmaking. In Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency (Athens, Greece) (FAccT '25) . Association for Computing Machinery, New York, NY, USA, 1640-1652. doi:10.1145/3715275.3732109 - source
    • cites [94] [94] Nari Johnson, Elise Silva, Harrison Leon, Motahhare Eslami, Beth Schwanke, Ravit Dotan, and Hoda Heidari. 2025. Legacy Procurement Practices Shape How U.S. Cities Govern AI: Understanding Government Employees' Practices, Challenges, and Needs. In Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency (Athens, Greece) (FAccT '25) . Association for Computing Machinery, New York, NY, USA, 772-789. doi:10.1145/3715275. 3732049 - source
  • Premisecitation
    § [Emma Kallina ∗](https://orcid.org/0000-0003-4912-7216) · p. 2

    The contractual terms and requirements central to procurement processes directly shape and govern the design of the procured system.

    “Moreover, the contractual terms and requirements that are central to procurement processes directly shape and govern the design of the procured system, further highlighting the relevance of public procurement for this research domain [9, 13, 95].”
    Quote in the paperSubstantial claim2 of 3 sources checked
    • cites [9] [9] Local Government Association. 2025. Responsibly Buying AI - How to build equality & data protection into your AI commissioning and procurement processes: a guide for councils in England . Technical Report. Local Government Association. -

      Not checked in the free run.

    • cites [13] [13] Dan Bateyko and Karen Levy. 2025. One Bad NOFO? AI Governance in Federal Grantmaking. In Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency (Athens, Greece) (FAccT '25) . Association for Computing Machinery, New York, NY, USA, 1640-1652. doi:10.1145/3715275.3732109 - source

      Section 5.1, “Lessons from AI Procurement Literature” (page unavailable), says contracting lets officials tailor restrictions to deployed AI tools and that procurement officials make policy choices about their purchase and use. It does not say that contractual requirements directly shape the procured system’s design. (LLM, conf=0.81)

      Experimental — accuracy not yet measured
    • cites [95] [95] Emma Kallina, Thomas Bohné, and Jatinder Singh. 2025. Stakeholder Participation for Responsible AI Development: Disconnects Between Guidance and Current Practice. In The 2025 ACM Conference on Fairness, Accountability, and Transparency (FAccT '25) (Athens, Greece) (FAccT '25) . Association for Computing Machinery, New York, NY, USA, to appear. doi:10.1145/3715275.3732069 - source

      Section 4.1 (pp. 6, 8) says early decisions shape downstream design and that business priorities can determine system objectives. It does not say that contractual terms or procurement requirements directly govern the design of a procured system. (LLM, conf=0.74)

      Experimental — accuracy not yet measured
  • May combine two claims. 49
  • Classified as probably not a substantial claim. 61
  • The quote only partly states this claim. 4

  • May combine two claims.

    Premisecitation
    § 2.2 Public sector AI: adoption, suppliers, and safeguards · p. 3

    AI could amplify existing challenges due to its technical complexity, increased autonomy, system opacity that impedes error detection, distributed responsibility across multiple suppliers and systems, and risks that propagate across linked systems and models.

    “While some of the discussed risks may also apply for traditional IT systems [61], AI could amplify existing challenges due to its technical complexity, increased autonomy, system opacity that impedes error detection, distributed responsibility across multiple suppliers and systems, and risks that propagate across linked systems and models [88, 91].”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites [88] [88] Sasonoto Hasbullah. 2025. AI Procurement Governance in Indonesian Customs: A Systematic Literature Review. doi:10.2139/ssrn.5356967 - source
    • cites [91] [91] Merve Hickok. 2024. Public procurement of artificial intelligence systems: new risks and future proofing. AI & SOCIETY 39, 3 (June 2024), 1213-1227. doi:10.1007/s00146-022-01572-2 - source

  • May combine two claims.

    Premisecitation
    § 2.3 Public procurement for AI governance: aspirations vs practice · p. 4

    The public sector rarely exercises its procurement power effectively; rigorous procurement processes are often bypassed through informal purchasing routes, and suppliers frequently withhold information or resist contractual changes.

    “Another US study confirms this trend, showing that the public sector rarely exercises its procurement power effectively: rigorous procurement processes are often bypassed through informal purchasing routes. Even when formal procurement takes place, suppliers frequently withhold information or resist contractual changes, which impacts the leverage of public authorities [94].”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [94] [94] Nari Johnson, Elise Silva, Harrison Leon, Motahhare Eslami, Beth Schwanke, Ravit Dotan, and Hoda Heidari. 2025. Legacy Procurement Practices Shape How U.S. Cities Govern AI: Understanding Government Employees' Practices, Challenges, and Needs. In Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency (Athens, Greece) (FAccT '25) . Association for Computing Machinery, New York, NY, USA, 772-789. doi:10.1145/3715275. 3732049 - source

      The Conclusion reports that most AI technologies in the studied cities were acquired through alternative pathways rather than formal solicitation. Section 5.2 reports that AI vendors frequently withheld information, resisted changes to contract terms, and left city employees feeling they lacked leverage. The citing sentence extends these findings about municipal AI procurement to the public sector generally, which this study does not establish. (LLM, conf=0.87)

      Experimental — accuracy not yet measured

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductcitation
    § 3.1 Materials and procedure · p. 5

    The selection process resulted in 16 interviews, a sample size consistent with prior exploratory research with hard-to-reach expert populations.

    “The selection process resulted in 16 interviews, a sample size consistent with prior exploratory research with hard-to-reach expert populations [see e.g. 4, 32, 154] (§7.5).”
    Quote in the paperNot a substantial claim3 of 3 sources checked
    • cites [4] [4] Michelle Adiwangsa, Hanna Suominen, and Penny Sweetser. 2025. Can Augmented Reality Head-Mounted Display Exergames Support the Management of Multiple Sclerosis at Home? Workshop Discussions with Researchers and Experts with Lived Experience of MS. In Proceedings of the Extended Abstracts of the CHI Conference on Human Factors in Computing Systems (CHI EA '25) . - Association for Computing Machinery, New York, NY, USA, Article 124, 7 pages. doi:10.1145/3706599.3720214 - source

      The Methodology describes a small study with eight researchers and experts with lived experience of MS, supporting the citation as an example of exploratory research with a specialist population. It reports a workshop, however, and does not describe its participants as hard to reach or discuss whether a sample of 16 interviews is consistent with prior work. The 16 interviews are the citing authors’ own result, not a number attributed to this source. (LLM, conf=0.78)

      Experimental — accuracy not yet measured
    • cites [32] [32] Yuexi Chen, Yimin Xiao, Kazi Tasnim Zinat, Naomi Yamashita, Ge Gao, and Zhicheng Liu. 2025. Comparing Native and Non-native English Speakers' Behaviors in Collaborative Writing through Visual Analytics. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems (CHI '25) . Association for Computing Machinery, New York, NY, USA, Article 1174, 16 pages. doi:10.1145/3706598.3713693 - source

      Section 7.1 (p. 9) describes a small qualitative evaluation involving four communication researchers and eight graduate students, so the paper offers a precedent for small-scale exploratory research. It does not report 16 interviews or characterize its participants as a hard-to-reach expert population; §7.1 describes brief interviews only after the eight individual study sessions. (LLM, conf=0.78)

      Experimental — accuracy not yet measured
    • cites [154] [154] Sirui Tao, Ivan Liang, Cindy Peng, Zhiqing Wang, Srishti Palani, and Steven P. Dow. 2025. DesignWeaver: Dimensional Scaffolding for Text-to-Image Product Design. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems (CHI '25) . Association for Computing Machinery, New York, NY, USA, Article 425, 26 pages. doi:10.1145/3706598.3714211 - source

      Section 3 (Formative Study; page unavailable) reports 12 interviews with experienced designers and says that sample size aligns with prior formative HCI studies, supporting the comparison to a 16-interview exploratory study. It does not characterize its expert participants as hard to reach, so it does not establish that part of the comparison. The 16 interviews are the citing study’s own count, not a count attributed to DesignWeaver. (LLM, conf=0.79)

      Experimental — accuracy not yet measured

  • Classified as probably not a substantial claim.

    Interpretationcitation
    § 7.3 Power asymmetries in AI procurement: towards empowering public sector buyers · p. 15

    HCI scholarship has a distinctive role in translating abstract principles into concrete, citizen-centred practices by offering insights into participatory design methods for AI systems and policy, tools to facilitate stakeholder engagement, and socio-technical evaluation.

    “HCI scholarship has a distinctive role here: by offering established insights into participatory design methods for AI systems and policy [87, 103, 183], tools to facilitate stakeholder engagement [97], and socio-technical evaluation [12, 110, 131], it helps translate abstract principles into concrete, citizen-centred practices.”
    Quote in the paperSubstantial claim3 of 7 sources checked
    • cites [87] [87] Christina Harrington, Sheena Erete, and Anne Marie Piper. 2019. Deconstructing community-based collaborative design: Towards more equitable participatory design engagements. Proceedings of the ACM on human-computer interaction 3, CSCW (2019), 1-25. - source

      Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured
    • cites [103] [103] Tzu-Sheng Kuo, Quan Ze Chen, Amy X. Zhang, Jane Hsieh, Haiyi Zhu, and Kenneth Holstein. 2025. PolicyCraft: Supporting Collaborative and Participatory Policy Design through Case-Grounded Deliberation. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems (CHI '25) . Association for Computing Machinery, New York, NY, USA, Article 805, 24 pages. doi:10. 1145/3706598.3713865 - source

      Split into 2 atoms — 1 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured
    • cites [183] [183] Douglas Zytko, Pamela J. Wisniewski, Shion Guha, Eric PS Baumer, and Min Kyung Lee. 2022. Participatory design of AI systems: opportunities and challenges across diverse users, relationships, and application domains. In CHI Conference on Human Factors in Computing Systems Extended Abstracts . 1-4. - source
    • cites [97] [97] Emma Kallina and Jatinder Singh. 2025. Mapping the Tool Landscape for Stakeholder Involvement in Participatory AI: Strengths, Gaps, and Future Directions. In Proceedings of the Extended Abstracts of the CHI Conference on Human Factors in Computing Systems (CHI EA '25) . Association for Computing Machinery, New York, NY, USA, Article 389, 8 pages. doi:10.1145/3706599.3719726 - source
    • cites [12] [12] Louise Barkhuus and Jennifer A Rode. 2007. From mice to men-24 years of evaluation in CHI. In Proceedings of the SIGCHI conference on Human factors in computing systems , Vol. 10. ACM New York, NY. - source
    • cites [110] [110] Yu Lu Liu, Wesley Hanwen Deng, Michelle S Lam, Motahhare Eslami, Juho Kim, Q Vera Liao, Wei Xu, Jekaterina Novikova, and Ziang Xiao. 2025. Humancentered evaluation and auditing of language models. In Proceedings of the Extended Abstracts of the CHI Conference on Human Factors in Computing Systems . 1-7. - source
    • cites [131] [131] Victor Ojewale, Ryan Steed, Briana Vecchione, Abeba Birhane, and Inioluwa Deborah Raji. 2025. Towards AI Accountability Infrastructure: Gaps and Opportunities in AI Audit Tooling. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems (CHI '25) . Association for Computing Machinery, New York, NY, USA, Article 815, 29 pages. doi:10.1145/3706598.3713301 - source

  • May combine two claims.

    Findingown data
    § 4.3 AI enters the public sector via additional channels, affording less oversight · p. 7

    Public sector buyers are procuring digital services that then get upgraded with AI capabilities, which may or may not be obvious.

    “Rather than explicitly 'procuring their AI technologies, [public sector buyers are][...] procuring a digital service that then gets upgraded with AI capabilities, which may or may not be so obvious' (P10) -a phenomenon often described as 'function creep' (P6).”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Findingown data
    § 4.1 Formal procurement of AI: folded into standard IT processes but AI-specific tenders are emerging · p. 6

    The maturity of AI-specific procurement in the UK and EU is still low.

    “Our participants reported that the maturity of AI-specific procurementi.e. public tenders that directly mention 'AI'-in the UK and EU is still low: 'it feels very basic [...] I don't even think we're there yet' (P13).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § Abstract · p. 1

    The study's findings result in six promising procurement practices that enable the public sector to shape AI in line with public interests, alongside concrete mechanisms to support their uptake.

    “Our findings result in six promising procurement practices that enable the public sector to shape AI in line with public interests, alongside concrete mechanisms to support their uptake.”
    • Also stated in: § [Emma Kallina ∗](https://orcid.org/0000-0003-4912-7216) · p. 2 · plain
      “Identifying six actionable promising practices and concrete mechanisms to realise these that directly support the public sector and its wider stakeholder network in leveraging its procurement power.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § Abstract · p. 1

    AI-specific procurement approaches remain immature, and formal AI-specific procurement appears to be in an early stage in the UK and EU.

    “Further, we find that AI-specific procurement approaches remain immature and systems often enter through informal channels with less scrutiny.”
    • Also stated in: § [Emma Kallina ∗](https://orcid.org/0000-0003-4912-7216) · p. 2 · hedged
      “and that formal, AI-specific procurement itself appears to be in an early stage in the UK and EU.”
    Quote in the paperSubstantial claim

  • May combine two claims. The quote only partly states this claim.

    Findingcitation
    § 4.2 AI procurement under framework contracts: privileged routes, unclear AI safeguards · p. 7

    Local public bodies can decide for themselves whether to adopt a system under a framework contract, but they don't always know what assessments were made on these providers.

    “Local public bodies 'can decide for themselves' (P15) whether to adopt a system under a framework contract; however, they”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § 4.1 Formal procurement of AI: folded into standard IT processes but AI-specific tenders are emerging · p. 6

    AI-specific tenders are emerging, but in an ad hoc manner.

    “While AI-specific tenders are emerging, this occurs in an ad hoc manner.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Findingown data
    § 4.2 AI procurement under framework contracts: privileged routes, unclear AI safeguards · p. 7

    Framework contracts allow public sector organisations to procure without launching a new tendering process, making it an easy way to obtain services.

    “This allows public sector organisations to procure without launching a new tendering process, meaning that 'they do not have to prepare a tender [...] [instead they can] simply ask [...] the operator within a sort of adhesion to the framework contract' , making it 'an easy way to obtain services' (P12).”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Findingown data
    § 4.2 AI procurement under framework contracts: privileged routes, unclear AI safeguards · p. 7

    Framework contracts lack mechanisms for rapid experimentation, outcome-based requirements, modular contracting, context adaptation, and iterative changes.

    “As P9 argued, 'the general approach to creating framework agreements is outdated and is not really geared up for the specifics and the nuances of AI procurement' since framework contracts lack mechanisms for rapid experimentation, outcome-based requirements, modular contracting, context adaptation, and iterative changes.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Findingown data
    § 5.1 Lack of clear guidance leads to weak embedding of public values and regulations · p. 8

    Traditional IT procurement methods were perceived as ill-suited to the dynamic and rapidly evolving nature of AI.

    “Additionally, traditional IT procurement methods were perceived as ill-suited to the dynamic and rapidly evolving nature of AI: 'if you procure with a process that's been done for 50 years, that way you can't be innovative' , which is important since 'with AI [...] you can't define an end product' (P7).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § 4.3 AI enters the public sector via additional channels, affording less oversight · p. 7

    There is shadow AI in a lot of offers; the procurement will not be categorised as AI, but there is AI software in it.

    “Indeed, P2 confirmed that 'there is shadow AI in a lot of offers. So the procurement will not be categorised as AI, but there is AI software in it' .”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Findingown data
    § 5.3 Strong vendor influence and skewed competition · p. 10

    Public procurement mechanisms usually lead us to big companies that don't have the technical expertise or motivation to meet those specific problems.

    “However, 'public procurement mechanisms usually lead us to big companies that don't have the technical expertise [or motivation] to meet those' (P12), 'forcing public sector buyers to rely on a few [big organisations] who may not deliver real value' (P10).”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data
    § 5.4 Complex infrastructures create vendor lock-in · p. 11

    Public sector organisations sometimes must buy back data they supplied to vendors to obtain necessary information for impact assessments.

    “This can take extreme forms, e.g. when 'public sector organisations sometimes must buy back data they supplied to vendors to obtain necessary information for impact assessments' (P10).”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Findingown data
    § 5.1 Lack of clear guidance leads to weak embedding of public values and regulations · p. 8

    Public interest-related aspects, especially around rAI, are currently rarely binding or impactful and instead treated as a 'nice-to-have'.

    “Related to the concerns around AI-related harms, P10 mentioned that public interest-related aspects (especially around rAI) are currently rarely binding or impactful and instead treated as a 'nice-to-have' .”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data
    § Figure 2: Visualisation of the relationships between codes and resulting themes describing key challenges as derived from our expert sample. · p. 10

    Unnecessary or unrealistic specifications may result in a complete waste of time and energy.

    “Unnecessary or unrealistic specifications may result in 'a complete waste of time and energy' (P5).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data
    § 5.1 Lack of clear guidance leads to weak embedding of public values and regulations · p. 8

    Participants pointed to challenges around selecting tender requirements and assessing compliance with them.

    “Resulting from this lack of actionable guidance, participants pointed to challenges around selecting tender requirements and assessing compliance with them.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § Figure 2: Visualisation of the relationships between codes and resulting themes describing key challenges as derived from our expert sample. · p. 10

    Such tenders risk irrelevance: some of the worst RFPs are those where it's not clear what they want but they're very detailed on how they want it done.

    “Such tenders risk irrelevance: 'some of the worst RFPs are those where it's not clear what they want [...] but they're very detailed on how they want it done' (P3), i.e. their focus is on legal or technical compliance instead of assessing how effective a system addresses a specific problem.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data
    § 5.4 Complex infrastructures create vendor lock-in · p. 11

    This can require access to data sources of suppliers, e.g. when a Data Protection Impact Assessment is to be conducted.

    “This can require access to data sources of suppliers, e.g. when a Data Protection Impact Assessment is to be conducted (P14).”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Findingown data
    § 6.2 Harness insights through knowledge sharing and collaboration in the public sector · p. 11

    Participants associated knowledge sharing with mitigating the current fragmentation and ad hoc procurement practices that risk wasting resources.

    “Participants associated such sharing with mitigating the current fragmentation and ad hoc procurement practices that risk wasting resources (§4.1).”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data
    § 7.5 Limitations and future work · p. 16

    Around half of the participants had experience spanning both current and former roles, providing both insider knowledge and reflective distance.

    “Around half had experience spanning both current and former roles, thus providing both insider knowledge and reflective distance.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Findingown data
    § 6.1 Provide clarity on vision, guidance, and operationalisation · p. 11

    Participants across countries emphasised that a central vision on AI must be actionable in the procurement practice.

    “Importantly, participants across countries emphasised that a central vision on AI must be actionable in the procurement practice, highlighting the need for guidance that clarifies how to operationalise such a vision .”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § 5.2 Limited organisational capacity & AI (procurement) literacy · p. 8

    AI literacy, capacity, and organisational structures as well as culture currently pose challenges for AI procurement.

    “Our sample emphasised that AI literacy, capacity, and organisational structures as well as culture currently pose challenges for AI procurement.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Findingown data
    § 5.1 Lack of clear guidance leads to weak embedding of public values and regulations · p. 8

    EU AI Act contractual clauses were widely seen as merely a starting point.

    “While several participants were aware of the existence of (or had even contributed to) the EU AI Act contractual clauses (§2.3.1), these were widely seen as merely a starting point.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Findingown data
    § 5.4 Complex infrastructures create vendor lock-in · p. 11

    Public sector datasets often lack quality, structure, or legal clarity.

    “First, public sector datasets often lack quality, structure, or legal clarity (P10).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § 5.2 Limited organisational capacity & AI (procurement) literacy · p. 8

    Buyers fear being tricked, are uncertain about which providers are reliable, and are sceptical about the often overhyped productivity benefits of AI.

    “This leaves buyers fearing 'being tricked' (P2), uncertain about which providers are reliable (P11) and sceptical about the 'often overhyped' productivity benefits of AI (P1).”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data
    § 5.4 Complex infrastructures create vendor lock-in · p. 11

    Especially when IT operations are outsourced, ownership of data and platform management complicates procurement conversations regarding AI solutions layered atop these systems.

    “Additionally and especially when IT operations are outsourced, 'ownership of data and platform management complicates procurement conversations regarding AI solutions layered atop these systems' (P11).”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Findingown data
    § 5.1 Lack of clear guidance leads to weak embedding of public values and regulations · p. 8

    There is no clear and consistent definition of what constitutes AI.

    “In this context, there is no clear and consistent definition of what constitutes AI: 'everything does have the name AI in it and so that obviously blurs the lines' (P2).”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Findingown data
    § 5.3 Strong vendor influence and skewed competition · p. 10

    Public authorities are in a conversation with different IT service providers, whose presence at industry-government forums influences how they write tenders.

    “Public authorities are in 'a conversation with different IT service providers' , whose presence at industry-government forums 'at the end of the day influence[s] how they write tenders' , though 'not a single IT service provider should be able to influence a specific tender' (P3).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § Figure 2: Visualisation of the relationships between codes and resulting themes describing key challenges as derived from our expert sample. · p. 10

    Few public bodies have dedicated digital procurement teams; most lack a technical focus.

    “Few public bodies have dedicated digital procurement teams; most lack a technical focus'buy post-its on Monday, AI on Tuesday' (P9)-and thus have 'no experts that know how to specify a solution' (P1).”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Findingown data
    § 6.2 Harness insights through knowledge sharing and collaboration in the public sector · p. 11

    Interviewees suggested that knowledge sharing, particularly the exchange of good practices, should take place at any level: local to national or transnational government, between local government bodies, within organisations, and between the public sector and research institutions.

    “Interviewees suggested that knowledge sharing, and particularly the 'exchange of good practices' , should take place 'on any level' (P11): local to national (or transnational) government, between local government bodies, within organisations, and between the public sector and research institutions.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data
    § 5.3 Strong vendor influence and skewed competition · p. 10

    This practice can stifle innovation and market competition by preventing bidders from proposing alternative solutions and by excluding smaller players.

    “Participants stressed that this practice can stifle innovation and market competition by preventing bidders from proposing alternative solutions ( 'fail to get some sort of innovation underway' ; P5) and by excluding smaller players without existing access or the resources for extensive marketing.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingnone given
    § 7.3 Power asymmetries in AI procurement: towards empowering public sector buyers · p. 15

    Public sector buyers often lack clear guidance, AI-specific expertise, and sufficient leverage, leaving them vulnerable to vendor influence and market hype rather than actively shaping procurement outcomes.

    “In contrast, public sector buyers often lack clear guidance, AIspecific expertise, and sufficient leverage, leaving them vulnerable to vendor influence and market hype rather than actively shaping procurement outcomes.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Findingown data
    § 5.1 Lack of clear guidance leads to weak embedding of public values and regulations · p. 8

    A major bottleneck to effective AI adoption in the public sector is the absence of a clear, consistent, and widely shared vision or guidance.

    “Several participants reported that a major bottleneck to effective AI adoption in the public sector is the absence of a clear, consistent, and widely shared vision or guidance.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Findingown data
    § 6.2 Harness insights through knowledge sharing and collaboration in the public sector · p. 11

    Participants suggested establishing hubs or libraries that bundle existing support and insights, and platforms that allow public sector buyers to share positive experiences and failures.

    “Various approaches were suggested to achieve this, including establishing hubs or libraries that bundle existing support and insights , as well as platforms that allow public sector buyers to share positive experiences and, importantly, 'safely share the failures' (P11).”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingown data
    § 7.5 Limitations and future work · p. 16

    Participants provided insights from different standpoints, spanning buyers, sellers, advisors, experts, and researchers, as well as various levels of local to cross-national government.

    “Additionally, our participants provided insights from different standpoints (spanning buyers, sellers, advisors, experts, and researchers) as well as various levels of local to cross-national government.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data
    § Figure 2: Visualisation of the relationships between codes and resulting themes describing key challenges as derived from our expert sample. · p. 10

    Procurement is often organisationally disconnected from the real demands and from those who understand use cases.

    “Further, procurement is often organisationally disconnected, i.e. 'a separate department relatively disconnected to the real demands' (P8) and from those who understand use cases.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingcitation
    § 7.5 Limitations and future work · p. 16

    The sample was slightly skewed towards the UK and the Netherlands, early hubs in European public-sector AI adoption, but also included perspectives from Italy, Germany, Belgium, Slovenia, and EU-level actors.

    “Our sample was slightly skewed towards the UK and the Netherlands-early hubs in European public-sector AI adoption [108, 128]-but also included perspectives from Italy, Germany, Belgium, Slovenia, and EU-level actors (Table 2).”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [108] [108] David Leslie. 2019. Understanding Artificial Intelligence Ethics and Safety: A Guide for the Public Sector. https://openroboethics.org/wp-content/uploads/ 2021/07/ORI-Foresight-into-Artificial-Intelligence-Ethics-Launch-V1.pdf. (Accessed on 01/04/2025). -

      Not checked in the free run.

    • cites [128] [128] Municipality of Amsterdam via Overheid. 2025. The Algorithm Register: Municipality of Amsterdam. https://algoritmes.overheid.nl/en/organisatie/gm0363/ gemeente-amsterdam. (Accessed on 02/09/2025). -

      Not checked in the free run.

  • May combine two claims.

    Findingown data
    § 5.4 Complex infrastructures create vendor lock-in · p. 11

    Participants widely pointed to considerations such as biased training data, insufficient awareness of data provenance, and data protection obligations.

    “Beyond these ownership issues, participants widely pointed to considerations such as biased training data, insufficient awareness of data provenance, and data protection obligations.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data
    § 4.3 AI enters the public sector via additional channels, affording less oversight · p. 7

    Only contracts above a certain value require full procurement processes with extended bidding periods, monitoring, and review, whereas below this threshold, the process is less rigorous.

    “Only contracts above a certain value require full procurement processes with extended bidding periods, monitoring, and review, whereas below this threshold, 'the process [is] less rigorous' (P2) -a fact that is also exploited through supplier offers of 'free experimentation, quote-unquote, free prototypes' (P2).”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Findingown data
    § 5.4 Complex infrastructures create vendor lock-in · p. 11

    It remains unclear which companies have access to public body data and how they utilise it after development.

    “However, this is rarely considered: P10 observed that 'it remains unclear which companies have access to [public body] data and how they utilise it after development' .”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S1 (asserted).

    Findingown data
    § 5.1 Lack of clear guidance leads to weak embedding of public values and regulations · p. 8

    Certain fundamental rules that apply equally to AI as to everyday items are seen as far too agnostic and insufficient to address AI's specific characteristics.

    “Certain fundamental rules (§2.1) that apply equally to AI as to everyday items (e.g. chairs or pencils) are seen as 'far too agnostic' and insufficient to address AI's specific characteristics.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data
    § 5.3 Strong vendor influence and skewed competition · p. 10

    Large firms employ dedicated procurement staff with the skills and resources to handle the bureaucracy of complex tenders, whereas startups don't have this capacity, so big firms have better chances of winning.

    “Large firms employ 'dedicated procurement staff' (P4) with the skills and resources to handle the bureaucracy of complex tenders, whereas 'startups don't have this capacity, so big fi rms have better chances of winning' (P4).”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Findingown data
    § 5.3 Strong vendor influence and skewed competition · p. 10

    Large technology firms enjoy structural, capacity-related advantages that put them ahead and further entrench their market dominance.

    “Within this dynamic, participants stressed that large technology firms enjoy structural, capacity-related advantages that put them ahead and further entrench their market dominance.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Findingnone given
    § 8 Conclusion · p. 16

    AI-specific procurement practices are currently fragmented and nascent, impeding effective scrutiny of the AI systems that enter public institutions.

    “AI-specific procurement practices are currently fragmented and nascent, impeding effective scrutiny of the AI systems that enter public institutions.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Findingown data
    § 5.1 Lack of clear guidance leads to weak embedding of public values and regulations · p. 8

    Procurers and different actors in the system are buying these technologies without really knowing what the technology is and what it does.

    “Consequently, 'procurers, different actors in the system, are buying these technologies without really knowing, one, what the technology is, and two, what it does' (P10).”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Findingown data
    § 6.2 Harness insights through knowledge sharing and collaboration in the public sector · p. 11

    Sharing knowledge, experiences, and AI systems within and across organisations was reported by participants to be crucial for improving public sector AI procurement and deployment.

    “Sharing knowledge, experiences, and AI systems-both within and across organisations-was reported to be crucial for improving public sector AI procurement and deployment.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given
    § 3.1 Materials and procedure · p. 5

    This work was approved by the department's ethics review board.

    “This work was approved by our department's ethics review board.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductcitation
    § 3.1 Materials and procedure · p. 5

    Before each interview, the working definition of AI was clarified to the participant, drawing on the EU AI Act but extended to include rule-based models, Internet of Things devices, and data infrastructures/platforms that interact with AI.

    “Before each interview, we clarified our working definition of AI to the participant, drawing on the EU AI Act [3], but extending it in this study to also include rule-based models, Internet of Things devices, and data infrastructures/platforms that interact with AI.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [3] [3] 2024. Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). Official Journal of the European Union, L 1689, 12 July 2024. -

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Study design and conductreasoning
    § 3.1 Materials and procedure · p. 5

    The study deliberately cast a wide net to include perspectives on AI procurement across different roles and government agencies, ensuring a broad range of views rather than a single organisational perspective.

    “Given the exploratory nature of the study, we deliberately cast a wide net to include perspectives on AI procurement across different roles and government agencies, ensuring a broad range of views rather than a single organisational perspective.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given
    § 3.1 Materials and procedure · p. 5

    All interviews were conducted by at least one lead researcher.

    “All interviews were conducted by at least one lead researcher.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductreasoning
    § 3.1 Materials and procedure · p. 5

    To ensure useful insights into public sector AI procurement, participants were required to have professional expertise in the procurement of digital products and services.

    “To ensure that our sample had useful insights into public sector AI procurement (as opposed to other forms of public sector procurement) we required participants to have professional expertise in the procurement of digital products and services.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductown data
    § 3.1 Materials and procedure · p. 5

    The sample included experts in public sector AI procurement: public sector buyers, authors of procurement guidance, technology suppliers bidding for public tenders, lawyers, procurement researchers, and both public and private consultants.

    “Our sample included experts in the emerging domain of public sector AI procurement: public sector buyers, authors of procurement guidance, technology suppliers bidding for public tenders, lawyers, procurement researchers, and both public and private consultants.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given
    § 3.1 Materials and procedure · p. 5

    Most interview sessions were attended by both lead researchers to ensure shared awareness, consistency in the interview process, and timely discussion of uncertainties.

    “Most sessions were attended by both lead researchers to ensure shared awareness, consistency in the interview process, and timely discussion of uncertainties as they emerged.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductown data
    § 3.1 Materials and procedure · p. 6

    The qualitative data were analysed using thematic analysis with independent coding by two researchers and consensus resolution to increase independence and objectivity.

    “After de-identifying our participants (P1-P16), the two lead researchers analysed the qualitative data following the six phases of a thematic analysis approach [21]. To increase independence, the first three phases were initially completed separately by both researchers, i.e. they coded several interviews independently to derive first sub-themes. Then, they compared codes and sub-themes, resolving”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [21] [21] Virginia Braun and Victoria Clarke. 2006. Using thematic analysis in psychology. Qualitative Research in Psychology 3, 2 (Jan. 2006), 77-101. doi:10.1191/ 1478088706qp063oa - source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given
    § 3.1 Materials and procedure · p. 5

    The sample reflected experiences at different government levels, ranging from local to trans-national (EU) levels.

    “The sample reflected experiences at different government levels, ranging from the more local (i.e. municipality/council) to trans-national (i.e. EU) levels.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation
    § 2.3 Public procurement for AI governance: aspirations vs practice · p. 4

    In Indonesia, AI procurement in customs was more complex than traditional IT but lacked tailored governance frameworks to enhance transparency and accountability.

    “In Indonesia, AI procurement in customs was found to be more complex than traditional IT but lacked tailored governance frameworks to enhance transparency and accountability [88].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [88] [88] Sasonoto Hasbullah. 2025. AI Procurement Governance in Indonesian Customs: A Systematic Literature Review. doi:10.2139/ssrn.5356967 - source

  • May combine two claims.

    Premisecitation
    § 2.2 Public sector AI: adoption, suppliers, and safeguards · p. 4

    Risk-prediction systems in child welfare services reproduced structural biases in administrative data and required extensive fairness auditing to assess their impact.

    “This is illustrated by risk-prediction systems in child welfare services that reproduced structural biases in administrative data and required extensive fairness auditing to assess their impact [36].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [36] [36] Alexandra Chouldechova, Diana Benavides-Prado, Oleksandr Fialko, and Rhema Vaithianathan. 2018. A case study of algorithm-assisted decision making in child maltreatment hotline screening decisions. In Proceedings of the 1st Conference on Fairness, Accountability and Transparency (Proceedings of Machine Learning Research, Vol. 81) , Sorelle A. Friedler and Christo Wilson (Eds.). PMLR, 134-148. https://proceedings.mlr.press/v81/chouldechova18a.html -

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Study design and conductnone given
    § 3.1 Materials and procedure · p. 5

    Participants provided informed consent.

    “Participants provided informed consent.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductnone given
    § 3.1 Materials and procedure · p. 5

    Participation was voluntary and no compensation was offered.

    “Participation was voluntary and no compensation was offered.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductown data
    § 3.1 Materials and procedure · p. 5

    All participants had in-depth experience with digital procurement and AI-specific practices, confirmed through a short survey and at the start of each interview.

    “Indeed, all participants not only had in-depth experience with digital procurement but also with AI-specific practices, as confirmed through a short survey as well as at the start of each interview.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning
    § 4.2 AI procurement under framework contracts: privileged routes, unclear AI safeguards · p. 7

    The result is a growing presence of AI in public systems that favours large, established suppliers, without clear AI-specific safeguards.

    “The result is a growing presence of AI in public systems that favours large, established suppliers, again without clear AI-specific safeguards.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning
    § 8 Conclusion · p. 16

    Public procurement is a potentially powerful but underutilised lever for governing how AI is adopted by the public sector.

    “This work shows that public procurement is a potentially powerful but underutilised lever for governing how AI is adopted by the public sector.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning
    § 8 Conclusion · p. 16

    There is a critical role for the human-centred research communities in supporting public-interest-aligned procurement practices.

    “We thereby identify a critical role for the human-centred research communities in supporting public-interest-aligned procurement practices and outline urgent directions for future research.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data
    § Figure 2: Visualisation of the relationships between codes and resulting themes describing key challenges as derived from our expert sample. · p. 10

    Accountability pressures reinforce this: no one wants to take the responsibility to decide professionally, so compliance to formal requirements feels safer and easier to evaluate than judging whether a system truly addresses a real-world problem.

    “Accountability pressures reinforce this: 'no one wants to take the responsibility to decide professionally' (P14), so compliance to formal requirements feels safer and easier to evaluate than judging whether a system truly addresses a real-world problem.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationown data
    § 6.2 Harness insights through knowledge sharing and collaboration in the public sector · p. 11

    Collaborative platforms and networks for pooling expertise and demand could enable more coordinated procurements.

    “Further, collaborative platforms and networks for pooling expertise and demand could enable more coordinated procurements, such as in the case where 'users of drones across government-from fi re brigades to border police to water authorities-worked together to publish a document aimed at suppliers and procurement professionals' (P15).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning
    § 7.4 Clear opportunities for HCI scholarship · p. 15

    There is a pressing opportunity for the HCI community to introduce, harness, and evolve established HCI methods and scholarship to advance public sector AI procurement.

    “Importantly, we uncover a pressing opportunity for the HCI community to introduce, harness, and evolve established HCI methods and scholarship to advance public sector AI procurement.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning
    § 5.3 Strong vendor influence and skewed competition · p. 10

    These dynamics contribute to strong vendor influence and skewed competition in AI procurement.

    “Collectively, these dynamics contribute to strong vendor influence and skewed competition in AI procurement.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).

    Interpretationreasoning
    § 8 Conclusion · p. 16

    These insights may also have broader relevance in positioning procurement as a general governance mechanism for helping to align emerging technologies with the public interest.

    “We further suggest that these insights may also have broader relevance in positioning procurement as a general governance mechanism for helping to align emerging technologies with the public interest.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductcitation
    § 7.5 Limitations and future work · p. 16

    As a qualitative, exploratory study among a relatively niche and difficult-to-reach group of experts, the findings are not statistically generalisable.

    “As a qualitative, exploratory study among a relatively niche and difficult-to-reach group of experts, the findings are not statistically generalisable [156] but are likely to offer useful insights from real-world practice.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [156] [156] Steven Tenny, Janelle M. Brannan, and Grace D. Brannan. 2025. Qualitative Study. In StatPearls [Internet] . StatPearls Publishing, Treasure Island, FL. https: //pubmed.ncbi.nlm.nih.gov/29262162/ Accessed: 2026-02-05. -

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). The quote only partly states this claim.

    Study design and conductreasoning
    § A Interview protocol · p. 22

    The semistructured interview protocol's strategy of starting with broad, unprompted questions and probing specific themes later in each area of investigation supported completeness of the interviews.

    “This strategy allowed us to understand which aspects were salient for our interviewees while also supporting completeness.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning
    § 7.3 Power asymmetries in AI procurement: towards empowering public sector buyers · p. 15

    Measures such as problem-based procurement, design contests, and AI literacy efforts not only support more balanced negotiations but also build the capacity to demand AI-specific safeguards.

    “Such measures not only support more balanced negotiations but also build the capacity to demand AI-specific safeguards.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims. The quote only partly states this claim.

    Study design and conductreasoning
    § 7.5 Limitations and future work · p. 16

    The lead researchers' prior professional interactions facilitated access and contextual knowledge but also risked bias.

    “This facilitated access and contextual knowledge but also risked bias.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Interpretationreasoning
    § 7.3 Power asymmetries in AI procurement: towards empowering public sector buyers · p. 15

    Problem-based procurement grounded in citizen needs, design contests or hackathons that open opportunities for SMEs, and efforts to strengthen AI literacy and procurement skills can help public buyers engage more critically with vendors.

    “Problem-based procurement grounded in citizen needs, design contests or hackathons that open opportunities for SMEs, and efforts to strengthen AI literacy and procurement skills can help public buyers engage more critically with vendors.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationreasoning
    § 7.5 Limitations and future work · p. 16

    The findings are expected to be relevant for debates internationally, even as specific dynamics may differ.

    “Although our sample is European, many of the issues and practices raised are shared across public-sector procurement more broadly, and we expect our findings to be relevant for debates internationally, even as specific dynamics (e.g. in the US) may differ.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). The quote only partly states this claim.

    Study design and conductreasoning
    § A Interview protocol · p. 22

    The semistructured interview protocol's strategy of starting with broad, unprompted questions and probing specific themes later in each area of investigation allowed the researchers to identify which aspects were salient to interviewees.

    “This strategy allowed us to understand which aspects were salient for our interviewees while also supporting completeness.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationown data
    § 6.1 Provide clarity on vision, guidance, and operationalisation · p. 11

    A clear strategy could cut through diverging attitudes and clarify the line between beneficial and potentially harmful applications of AI.

    “A clear strategy could cut through such diverging attitudes and clarify the line between beneficial and potentially harmful applications of AI: 'We just need to have a clear modus operandi internally on like, OK, what do we want to do? What do we not want to do?' (P2).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning
    § 6.2 Harness insights through knowledge sharing and collaboration in the public sector · p. 11

    Knowledge sharing helps municipalities understand what already exists, reducing duplication.

    “Such sharing helps municipalities understand what already exists, reducing duplication.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning
    § 7.5 Limitations and future work · p. 16

    Many of the issues and practices raised are shared across public-sector procurement more broadly.

    “Although our sample is European, many of the issues and practices raised are shared across public-sector procurement more broadly, and we expect our findings to be relevant for debates internationally, even as specific dynamics (e.g. in the US) may differ.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given
    § 7.5 Limitations and future work · p. 16

    Both lead researchers had prior professional interactions with the public sector through civil society and government roles.

    “Both lead researchers had prior professional interactions with the public sector through civil society and government roles.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data
    § 4.3 AI enters the public sector via additional channels, affording less oversight · p. 7

    The fact that AI is being inserted as a new layer raises issues including uncertainties around the limited testing, assessment, or even awareness of AI updates.

    “In all, the fact that 'AI is being inserted as a kind of new layer [...] raises all sorts of issues' (P15), including uncertainties around the limited testing, assessment, or even awareness of AI updates.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning
    § 5.3 Strong vendor influence and skewed competition · p. 10

    Systems are often selected from what is hyped or well-promoted, making the relationship hugely skewed towards tech providers.

    “As a result, systems are often selected from what is hyped or well-promoted, prompting P10 to conclude that the relationship is 'hugely skewed towards tech providers' .”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning
    § 7.3 Power asymmetries in AI procurement: towards empowering public sector buyers · p. 15

    The findings underscore the need to actively rebalance power in AI procurement by strengthening the capacity, leverage, and agency of public sector buyers.

    “Taken together, these findings underscore the need to actively rebalance power in AI procurement by strengthening the capacity, leverage, and agency of public sector buyers.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning
    § 4.3 AI enters the public sector via additional channels, affording less oversight · p. 7

    Pilot projects are often structured to remain below these financial thresholds, which can reduce due-diligence requirements and allow systems to enter public services under reduced scrutiny.

    “Pilot projects are often structured to remain below these financial thresholds, which can reduce due-diligence requirements and allow systems to enter public services under reduced scrutiny.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S1 (asserted), S3 (load-bearing).

    Interpretationreasoning
    § 5.4 Complex infrastructures create vendor lock-in · p. 11

    They might train models with this data to sell it to other public sector buyers.

    “For example, they might train models with this data to sell it to other public sector buyers: 'controlling how AI learns from public data is an unresolved challenge' (P14).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationreasoning
    § 7.5 Limitations and future work · p. 16

    The findings are likely to offer useful insights from real-world practice.

    “As a qualitative, exploratory study among a relatively niche and difficult-to-reach group of experts, the findings are not statistically generalisable [156] but are likely to offer useful insights from real-world practice.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationown data
    § 6.1 Provide clarity on vision, guidance, and operationalisation · p. 11

    These findings highlight the need for a central AI vision that is concretely operationalised through procurement.

    “These findings highlight the need for a central AI vision that is concretely operationalised through procurement.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning
    § 5.4 Complex infrastructures create vendor lock-in · p. 10

    This blurs procurement boundaries and makes switching vendors even harder.

    “This blurs procurement boundaries and makes switching vendors even harder.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning
    § 7.5 Limitations and future work · p. 16

    The sample composition implies a good level of geographical diversity in perspectives across the EU and UK.

    “This implies a good level of geographical diversity in perspectives across the EU and UK.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationown data
    § 6.1 Provide clarity on vision, guidance, and operationalisation · p. 11

    This would help to support strategic alignment and coherent processes across government agencies, particularly given the uncertainty surrounding responsible AI policies in political arenas.

    “This would help to support strategic alignment and coherent processes across government agencies, particularly given the uncertainty surrounding responsible AI policies in political arenas (as discussed in §5.1).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning
    § 5.4 Complex infrastructures create vendor lock-in · p. 10

    While vendor lock-in is a known challenge in IT procurement, this seems especially pronounced in or exacerbated by the AI context.

    “Indeed, while vendor lock-in is a known challenge in IT procurement, this seems especially pronounced in or exacerbated by the AI context.”
    Quote in the paperNot a substantial claim
93 of 93

  • The paper relies on this without support.

    The accounts provided by the interview participants accurately reflect actual public sector AI procurement practices.

    “Our participants reported that the maturity of AI-specific procurementi.e. public tenders that directly mention 'AI'-in the UK and EU is still low”

  • The paper relies on this without support.

    The sample of 16 experts is sufficient to identify the key challenges and promising practices of public sector AI procurement.

    “After de-identifying our participants (P1-P16), the two lead researchers analysed the qualitative data following the six phases of a thematic analysis approach [21].”
2 of 2

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Source accessed before its dated publication

    Minor

    The citation's URL dates the Guardian article to 22 July 2025, but the reference says it was accessed on 20 July 2025, two days earlier.

1 of 1

Cited works that were retracted

Works this paper cites that were later retracted, from the retraction notices publishers deposit. Major when the notice came out before this paper was published.

  • Cites a retracted work: 10.1007/s13132-024-01929-6

    Major

    The cited work was retracted on 2025-11-19 (notice 10.1007/s13132-025-02956-7), before this paper was published (2026-04-13). Its authors could have known.

    Cited work 10.1007/s13132-024-01929-6 · retracted 2025-11-19 · notice 10.1007/s13132-025-02956-7 · this paper published 2026-04-13

1 of 1

Cites

1 retracted

Flagged first, then by citations

Research works printed but not linked: 17

  1. 2014. Directive 2014/24/EU on Public Procurement and Repealing Directive 2004/18/EC. Official Journal of the European Union L 94 28 March 2014 pp. 65–242. Entered into force on 17 April 2014.
  2. 2024. Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). Official Journal of the European Union L 1689 12 July 2024.
  3. Local Government Association. 2025. Responsibly Buying AI - How to build equality & data protection into your AI commissioning and procurement processes: a guide for councils in England. Technical Report. Local Government Association.
  4. Cloe Autio, Kate Cummings, Brinson S. Elliott, and Beth Simone Noveck. 2023. A Snapshot of Artificial Intelligence Procurement Challenges: Diagnosing perceived and actual risks impeding responsible AI acquisition in government. Technical Report. The GovLab. https://files.thegovlab.org/a-snapshot-of-ai-procurement-challenges-june2023.pdf
  5. 10.2799/0409819 · DOI
  6. Carine Lallemand and Vincent Koenig. 2017. Lab testing beyond usability: challenges and recommendations for assessing user experiences. Journal of Usability Studies 12 3 (2017).
  7. 10.1145/3313831.3376727 · DOI · Paper page
  8. Mark H. Moore. 1995. Creating Public Value: Strategic Management in Government. Harvard University Press, Cambridge, MA.
  9. 10.1787/ba7c7226-en · DOI
  10. National Audit Office. 2025. Government’s approach to technology suppliers: addressing the challenges. Session 2024-2025. Technical Report HC 543. NAO.
  11. UN OHCHR. 2025. Artificial intelligence procurement and deployment: ensuring alignment with the Guiding Principles on Business and Human Rights - Report of the Working Group on the issue of human rights and transnational corporations and other business enterprises. Technical Report A/HRC/59/53. https://www.ohchr.org/en/documents/thematic-reports/ahrc5953-artificial-intelligence-procurement-and-deployment-ensuring
  12. 10.1145/3706598.3713301 · DOI · Paper page
  13. Charlie Parker Sam Scott and Alistair Geddes. 2019. Snowball sampling. SAGE research methods foundations (2019).
  14. Leevi Saari. 2025. The Big Tech Playbook. IPS Journal (17 11 2025). https://www.ips-journal.eu/work-and-digitalisation/the-big-tech-playbook-8681/
  15. Steven Tenny, Janelle M. Brannan, and Grace D. Brannan. 2025. Qualitative Study. In StatPearls [Internet]. StatPearls Publishing, Treasure Island, FL. https://pubmed.ncbi.nlm.nih.gov/29262162/ Accessed: 2026-02-05.
  16. Lin Zhu Chiyu Chao and Zhiyong Fu. 2024. How HCI Integrates Speculative Thinking to Envision Futures. Journal of Futures Studies 28 4 (2024).
  17. Shoshana Zuboff. 2019. The Age of Surveillance Capitalism: The Fight for a Human Future at the New Frontier of Power. Pofile Books Ltd, London.

Cited by

Sorted by citations

Related

Authors
Hudig, A. I., Kallina, E., & Singh, J.
Affiliations
  • University of Cambridge (GB)
  • University of Duisburg-Essen (DE)
ORCID
Venue
Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems
Publisher
ACM
Published
2026-04-13
Pages
1-22
DOI
10.1145/3772318.3791968
OpenAlex
W7154147241
Cited by
3
References
130
Citations by year
2026: 3
FWCI
99.48
Type
conference paper
Language
en
Open access
Licence
Locations
  • CHI Conference on Human Factors in Computing Systems (publishedVersion, cc-by)
  • Apollo (University of Cambridge) (submittedVersion, cc-by)
  • Universitätsbibliographie, Universität Duisburg-Essen (submittedVersion)
Topics
  • Social Sciences › Business, Management and Accounting › Strategy and Management › Public Procurement and Policy
  • Social Sciences › Political Science and International Relations › E-Government and Public Services
  • Social Sciences › Safety Research › Ethics and Social Impacts of AI
Keywords
public procurement, AI governance, public sector AI, procurement practices, AI regulation, public interest

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-30.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-30.

Cited sources are checked by gpt-6-sol.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review