AI in procurement procedures: what contracting authorities, bidders, and advisers may do lawfully now

Artificial intelligence is entering procurement procedures—on contracting authorities' side as much as bidders'. Legally, supportive use is largely permitted. Once AI substantially decides the outcome, it may qualify as a high-risk system under the AI Act. Anyone wishing to exclude AI-generated concepts or require disclosure must state that clearly in the tender documents.

AI has arrived in procurement law

Systems such as ChatGPT, Copilot, Gemini, or Claude are part of everyday work and do not stop at procurement. Pilot projects run at federal and state level; AI-supported services are meant to shorten procedure times. A joint platform for data-based, networked procurement is also planned. There is also disillusionment: early trials in administration showed that systems tested for drafting procedures often still fall short.

A sober view is therefore sensible. Technology's benefits can be used without ignoring risks. What matters is distinguishing roles—whether AI is used by the contracting authority, bidder, in advisory work, or in review bodies. Different as these roles are, one requirement applies to all: without a minimum understanding of the technology, AI cannot be used sensibly or lawfully. That is where the AI Act bites.

AI competence is not optional but mandatory

Since the AI Act (Regulation (EU) 2024/1689) applies, Article 4 requires providers and deployers of AI systems to ensure, to the best of their ability, sufficient AI competence of their personnel—technical knowledge, experience, training, and awareness of context. Anyone using AI in procurement should therefore practise handling it deliberately.

In practice that mainly means precise formulation. Result quality depends directly on instruction precision: assign the AI a role and target audience, provide context with relevant norms and model texts, state formal requirements clearly, and break complex tasks into steps. Every output still needs independent legal and factual review.

Bidders' AI-generated text: permitted but hard to detect

It used to be assumed that submitted text came intellectually from the bidder. That premise no longer holds in the AI era. AI-generated text in procurement is generally permitted unless expressly excluded, and use alone is not attempted deception.

There is a practical problem: revised AI text is hard or impossible to detect. Detection tools are unreliable, especially with mixed texts. Typical indicators—a very smooth style, consistent structure, no typos, or consistently positive portrayal—may suggest AI but prove nothing, as each can appear in human text too. Proof is possible only in a reasoned case-by-case review to be documented under Section 8 VgV.

Why purely AI concepts are problematic for evaluation

Concepts are requested to assess know-how, capability, and innovation potential. A pure AI concept does not do that: what would be evaluated is not the bidder's capability but the program's—and because AI only draws on existing material, similar systems with similar prompts yield similar results, eroding discriminatory value.

That creates a design task for contracting authorities. Because reliably detecting AI text is practically impossible, the main response is the clarification interview under Section 15(5) VgV to test whether the bidder has intellectually grasped the task. It also helps to state in tender documents in advance that concepts created solely by AI are insufficient, or to oblige bidders to disclose AI use. Breach of a prior disclosure duty can lead to exclusion.

AI on the contracting authority side: supportive yes, high-risk only exceptionally

Supportive AI use by contracting authorities is already permitted. For example, AI under human supervision may speed formal bid review under Sections 56–57 VgV or help check bidder-related exclusion grounds under Sections 123–125 GWB by evaluating self-declarations or register extracts.

The decisive line is the risk concept in the AI Act. If AI were used comprehensively and outcome-determinatively for bid review—not merely supportively—it might qualify as high-risk under Article 6(2) with Annex III. The carve-out in Article 6(3) applies if the system does not pose significant risk to health, safety, or fundamental rights and does not substantially influence the decision outcome. Case-by-case value judgments are what AI currently cannot perform. As long as it remains a genuine auxiliary function—for example capturing direct awards—a high-risk system is remote.

What this means in practice

AI will not spare procurement law, and banning AI-generated documents would be neither sensible nor practicable. The main value of AI in procurement today lies in post-hoc review of text, while standalone drafting is limited by hallucinations. All parties benefit from building AI competence with realistic expectations.

Concretely: contracting authorities should clarify in tender documents whether AI concepts are allowed and must be disclosed, and plan clarification interviews. Bidders should use AI to refine their own thought-through concepts, not replace their own performance.