● LIVEECB RATE 2.4%·EUR/USD 1.1539·EUR/GBP 0.8558·BRUSSELS 21°C — OVERCAST·UKRAINE RAMPS UP REFINERY STRIKES AS TRUMP'S CEASE‑FIRE CLAIM PROVES UNFOUNDED·EUROPE FACES FINANCIAL, ECONOMIC AND DIGITAL SHOCKS THAT THREATEN ITS SECURITY·WORKERS AND UNIONS MUST SHAPE EUROPE'S NEW CIRCULAR ECONOMY LAW·EU COMPETITIVENESS FUND SPARKS US WARNING OVER 'MADE IN EUROPE' RULES·VON DER LEYEN IS EXPECTED TO TIE COMPETITIVENESS TO EUROPEAN STRATEGIC AUTONOMY IN STATE OF THE UNION·AI-GENERATED CONTENT OVERWHELMS WEB, PROMPTING GRASSROOTS ANTI‑BOT TOOLS·WEEK 38 · VOL. XV · N°258·● LIVEECB RATE 2.4%·EUR/USD 1.1539·EUR/GBP 0.8558·BRUSSELS 21°C — OVERCAST·UKRAINE RAMPS UP REFINERY STRIKES AS TRUMP'S CEASE‑FIRE CLAIM PROVES UNFOUNDED·EUROPE FACES FINANCIAL, ECONOMIC AND DIGITAL SHOCKS THAT THREATEN ITS SECURITY·WORKERS AND UNIONS MUST SHAPE EUROPE'S NEW CIRCULAR ECONOMY LAW·EU COMPETITIVENESS FUND SPARKS US WARNING OVER 'MADE IN EUROPE' RULES·VON DER LEYEN IS EXPECTED TO TIE COMPETITIVENESS TO EUROPEAN STRATEGIC AUTONOMY IN STATE OF THE UNION·AI-GENERATED CONTENT OVERWHELMS WEB, PROMPTING GRASSROOTS ANTI‑BOT TOOLS·WEEK 38 · VOL. XV · N°258·
Brussels · Est. 2012
Independent European journalism

Unionpress

Vol. XV · N°258
Tuesday, 15 September 2026
Home/Business
UNIONPRESSBusiness12 September 2026

Public‑sector procurement guidance on agentic AI remains unclear

Evidence that public buyers must compare the methods and outcomes of agentic AI tools, rather than rely on brochure…

Server rack hosting the agentic AI system in an Ontario government data centre
Server rack hosting the agentic AI system in an Ontario government data centre

Public buyers looking at agentic state AI should note that the evidence base for comparing methods and results remains thin.

What the available sources say

The Elsewhen report explains the concept of agentic AI and its potential to reshape public‑service workflows, from education to hospital discharge. It highlights time‑saving benefits but does not provide any procurement guidance or a requirement to evaluate tools beyond their marketing claims.

The UK Information Commissioner’s Office (ICO) Tech Futures paper discusses data‑protection implications of agentic AI and outlines four possible adoption scenarios. It stresses the need for organisations to understand technical capabilities and associated risks, yet it stops short of advising public buyers on how to compare methods or outcomes.

ReedSmith’s blog notes a recent uptick in regulatory activity, citing a Competition and Markets Authority (CMA) research paper on consumer protection. The CMA paper flags potential fines for misleading practices by AI agents but does not address procurement criteria for public‑sector contracts.

The GOV.UK publication provides a definition of agentic AI for consumers and lists everyday examples. It contains no mention of public‑sector procurement or evaluation frameworks.

Why the claim lacks verification

The commission’s brief centres on a claim that buyers must compare method and results rather than rely on brochure branding. The research packet explicitly states that “No verifiable evidence was found that buyers of agentic AI need to compare methods and outcomes instead of relying on brochure branding.” Consequently, the article cannot present the claim as established fact.

All four sources examined – Elsewhen, ICO, ReedSmith and GOV.UK – discuss the technology, its risks or its consumer impact, but none prescribe a specific procurement approach. Without a regulator‑issued guideline or a documented procurement policy, public‑sector buyers have no mandated requirement to perform the suggested comparison.

Implications for procurement teams

In the absence of explicit guidance, procurement officers should continue to apply existing public‑sector procurement rules. These include the EU Public Procurement Directive and the UK Public Contracts Regulations, which require a transparent assessment of value for money, sustainability and compliance with data‑protection law.

Teams can still adopt best‑practice measures: request demonstrators, obtain third‑party audit reports, and assess the alignment of AI agents with organisational risk frameworks. Such steps mirror the precautionary approach recommended by the ICO’s data‑protection outlook.

Where a supplier markets an agentic AI solution, buyers may ask for evidence of performance in comparable public‑service contexts. However, any such request must be proportionate and justified under the existing legal framework, as there is no separate statutory requirement to compare methods versus brochure branding.

What remains unknown

Future regulator publications may clarify whether a dedicated procurement standard will emerge. The CMA’s March 2026 paper on consumer protection does not extend to public‑sector contracts, but the agency’s broader mandate could evolve.

Stakeholders are also awaiting the outcome of the EU’s AI Act, which is expected to address high‑risk AI systems, including those used by public bodies. Until those rules are finalised, the procurement landscape for agentic AI will stay largely guided by general public‑procurement and data‑protection legislation.

Source coverage at a glance

Key sources and the aspects of agentic AI they address
Source Focus area Procurement guidance?
Elsewhen Conceptual overview, public‑service use cases No
ICO Data‑protection implications, scenario analysis No
ReedSmith Regulatory activity, consumer‑protection risks No
GOV.UK Definition and consumer context No
Source: research packet supplied to Union Press.

For further reading on procurement best practice, see Union Press’s recent analysis of ADHD test procurement Employers must assess ADHD test methods, not just brochure t and the guide on negotiation‑training procurement Beschaffung von Verhandlungsschulungen: Was Einkäufer jetzt . These pieces illustrate how public bodies can structure evidence‑based assessments even when sector‑specific guidance is absent.

■ ENDUnionPress© UnionPress 2026