Using AI to assist a procurement team is different from buying an AI system for public services. Both require clear responsibilities, reliable evidence and protection of confidential information. The intended use determines which additional rules apply.
AI as an administrative tool
Summaries and draft assessments must be checked against the original documents. An AI system must not introduce undisclosed award criteria or silently replace the authority’s reasoned decision. If it assists evaluation, the relevant reasoning and human review should remain traceable. Bidders likewise remain responsible for the accuracy of AI-assisted references, declarations and commitments.
Confidentiality and data handling
Before uploading tenders or calculations to a service, examine access, retention, training use, subcontractors and deletion. Removing a company name does not necessarily anonymise technical information or prices. Approved tools, permitted data and responsible reviewers should be specified internally. Procurement confidentiality and data protection apply independently of the timetable for high-risk AI obligations.
AI Act timetable in November 2026
The consolidated AI Act amended in July 2026 provides staged application dates. Central rules for Annex III high-risk systems generally apply from 2 December 2027; product-related high-risk systems under Article 6(1) generally follow on 2 August 2028. Certain transparency obligations already apply from August 2026. Each obligation, role and transitional rule must be checked separately.
Specify and test the intended use
Tender documents should define output quality, error limits, human intervention, logs, updates and exit. Tests need realistic and difficult scenarios. A general promise of AI Act compliance is not a substitute for classification or acceptance criteria. Model changes during performance may also raise contractual and procurement questions.
Four practical examples
The following scenarios are illustrative, not reports of actual client matters. The outcome depends on the documents and circumstances of the individual procurement.
Example 1: Summary tool
A team uses AI to summarise public documents. A reviewer checks every material statement against the originals. The generated summary is a working aid, not evidence that a requirement has been met.
Example 2: Automatic scoring
A tool proposes quality scores for tenders. The authority checks the published criteria, consistency and reasoning. An unexplained ranking cannot substitute for a properly documented evaluation.
Example 3: Invented experience
A bidder adopts an AI-generated description of a project it never delivered. Verify factual statements and supporting evidence before submission. Responsibility for the declaration remains with the bidder.
Example 4: Recruitment system
A public body buys software to assess job applicants. Its intended use may be important to high-risk classification. Procurement documents should address the applicable timetable, human oversight and evidence required from the supplier.
Related topics
IT and cloud procurement in Germany · Award criteria and tender evaluation · Contract modifications and additional public works
Discuss your procurement
For a review, please identify the authority, contract or lot, current procedural stage and any deadline. We can then clarify the scope of the requested work. Contact our team. An enquiry does not itself create a retainer or preserve a legal deadline.
Legal sources
Official legislation and materials, some in German: § 5 VgV – Vertraulichkeit · § 127 GWB – Zuschlag · KI-Verordnung – geltende Fassung · EU-Kommission: Art. 113 und Anwendungsfristen · EU-Kommission: Art. 4 – KI-Kompetenz
Updated 13 November 2026. Illustrative guidance on German and EU procurement law.