Executive Summary

State-subsidized AI training programs in Germany are increasingly being abused by disreputable providers. According to F.A.Z. research, many courses are superficial in content, and individual providers have been reported for subsidy fraud. Experts speak of "Corona test centers 2.0". The control structures of the accreditation and approval system (AZAV) have significant gaps – particularly in the technical quality assessment of AI content.

People

Topics

  • Subsidy fraud in training offerings
  • Lack of content quality in AI courses
  • Weaknesses in the AZAV certification system
  • Labor market policy reform needs

Clarus Lead

The market for state-subsidized AI training programs is fragmented and prone to abuse. The Federal Employment Agency distributed approximately 1.66 billion euros for vocational training in 2024. Research shows that individual providers abruptly end courses, lack genuine certifications, and commit subsidy fraud – without significant consequences. The Qualification Opportunities Act as an instrument for shaping digital structural change is discredited by inadequate oversight. Experts warn that a parallel market of low-quality offerings has developed that is "completely decoupled" from genuine AI training objectives.

Detailed Summary

The F.A.Z. documents a concrete case: A course provider advertised 100 percent state funding plus 75 percent wage cost reimbursement for a five-week AI expert course. Suddenly, the provider stopped the course and became unreachable – it turned out that the provider itself had no certification and only worked for a Saxony-Anhalt company. This company was instructed by the public prosecutor's office to cease all measures due to subsidy fraud. Such incidents are not isolated cases.

The content deficiencies are systemic. Many courses only cover ChatGPT operation but fail to address genuine business needs such as data management, model evaluation, and integration into corporate environments. The AI Federal Association criticizes lack of practical depth and has therefore founded its own GmbH to establish quality standards. Tristan Post, consultant for the UN Development Program, describes the situation sharply: providers "exploit the political momentum of AI, the funding logic of the Federal Employment Agency, and structural gaps in the Qualification Opportunities Act to sell programs that are neither quality-assured nor substantively credible."

The accreditation system fails for AI courses. While the AZAV (Accreditation and Approval Ordinance) regulates instructor qualifications and organizational criteria, it lacks depth in technological expertise. Auditors, themselves affected by AI skills shortages, often cannot technically assess content. An open secret in the industry: some auditors do not conduct on-site reviews but work from home offices. Participants, mostly without prior knowledge, cannot recognize quality deficiencies. In 2025, the Federal Employment Agency recognized initial problems and published warnings, but systematic reforms are only just beginning.

Key Findings

  • Fraud cases documented: Individual providers collect subsidies and discontinue courses without notice; the public prosecutor's office is investigating subsidy fraud.

  • Content superficiality: Many AI courses reduce themselves to ChatGPT operation and ignore real business needs such as data integration and model evaluation.

  • Systemic control deficiencies: The AZAV system checks formalities (fire safety, instructor qualifications) but not technical depth in AI content; auditors lack technological expertise.

  • Missing quality standards: Unlike established industries, there are no widely accepted training standards for AI competencies.

  • Mismatch between requirements and offerings: Small businesses often don't know what AI competencies they need; participants end up in unsuitable courses.

  • Reform needs recognized: The federal government and Federal Employment Agency plan modernization of the AZAV; examples like the United Kingdom show more centralized, transparent processes.

Critical Questions

  1. Evidence & Data Quality: How many subsidy fraud proceedings against AI course providers are currently pending, and what is the order of magnitude of financial damage? The Federal Employment Agency does not publish a breakdown of how the 1.66 billion euros are allocated to course types – are such transparency figures available?

  2. Conflicts of Interest: What financial dependency exists between certification bodies (TÜV, Dekra) and educational providers, and to what extent did this create pressure not to apply standards too strictly? Do auditors profit from quantity rather than quality of their inspections?

  3. Causality & Alternatives: Is it really the absence of quality standards that leads to disreputable offerings – or is it insufficient sanctions for fraud? Would strengthened on-site inspections without formal rule changes have been sufficient?

  4. Feasibility & Risks: A centralized British model is more efficient, but is it compatible with Germany's federal structures? Who bears the costs for modernizing the AZAV, and how long would a reform process take while participants continue to be enrolled in poor courses?

  5. Market Dynamics: Does state funding itself attract disreputable providers, or are these structural control deficiencies? Would a voucher system instead of flat-rate funding be more transparent?

  6. Feedback Structures: Why do local employment agencies receive no feedback from participants about course quality according to Marose, and how could this feedback mechanism be practically established?

Sources

Primary Source: The Great AI Training Scam – Frankfurter Allgemeine Zeitung, 17.02.2026

Supplementary Sources:

  1. Federal Ministry of Labor – Qualification Opportunities Act (QCG)
  2. Federal Employment Agency – Accreditation and approval of educational institutions
  3. German Accreditation Body GmbH (DAkkS)
  4. AI Federal Association – Positions on AI training
  5. Bitkom – Statement on digital training formats

Verification Status: ✓ 17.02.2026

This text was created with the support of an AI model. Editorial responsibility: clarus.news | Fact-checking: 17.02.2026