EU Commission breaches own AI guidelines by using ChatGPT in public documents (www.iccl.ie)

Ireland’s Digital Rights NGO ICCL has filed a complaint with the European Ombudsman after discovering that the European Commission used generative AI outputs in public documents — in one case revealed by a URL containing “utm_source=chatgpt.com.” The Commission’s disclosure came in response to an access-to-documents request and suggests at least one link was produced by OpenAI’s ChatGPT. ICCL argues this likely breaches the Commission’s own staff guidance, which forbids directly replicating generative-AI output in public documents, and may contravene treaty obligations for accurate administration. The case matters for the AI/ML community because it highlights institutional risks when public bodies adopt probabilistic language models without robust provenance, verification, and disclosure practices. Large language models predict next tokens and can produce fluent but incorrect “hallucinated” facts; that behavior is a design characteristic, not a bug. Implications include potential Ombudsman scrutiny, stricter documentation and audit requirements for AI-assisted drafting, and reinforced expectations that governments must disclose tool use and shoulder the burden of verification. The complaint could set a precedent for how transparency, liability and human-in-the-loop checks are enforced for generative AI in public-sector communications.