The European Parliament is fighting against "neural slop": a new platform for AI oversight among MEPs
The European Parliament has found an unconventional approach to the problem of uncontrolled use of neural networks among lawmakers. Instead of a total ban, the institution's leadership is launching its own platform, EPGenAI Hub, which will provide access to models from OpenAI, Meta, Anthropic, and France's Mistral. However, every request will now be under strict control.
The problem has been brewing for a long time. Texts generated by neural networks are increasingly infiltrating parliamentary work—from writing speeches and inquiries to amendments to laws. It is estimated that about 2,100 people, or nearly a fifth of the staff, use AI services daily. The situation reached its peak at the July meeting of committee chairs, when an anomalous surge in amendments requiring translation was recorded. The texts were characterized by a distinctly dry style and contained references to non-existent laws—classic "hallucinations" of language models.
Irish MEP Barry Andrews, who chairs the development committee, admitted that colleagues write inquiries using AI due to the enormous workload. However, for him personally, such tasks remain key responsibilities. For now, lawmakers are not required to disclose their use of neural networks.
European-Style Cat in the Soup
The EPGenAI Hub platform has already been presented to MEPs at workshops, with the official launch scheduled for September. The beta version allows users to choose between local models (Llama from Meta, GPT-OSS from OpenAI, Mistral Small) and external services (ChatGPT and Claude Sonnet). However, here lies the main paradox—by launching the platform, the parliament is expanding the presence of American AI giants amid Brussels' push to reduce technological dependence on the US. Previously, the institution replaced Google with the French search engine Qwant and disabled built-in AI on MEPs' tablets for cybersecurity reasons.
The parliament's press service clarified that the list of available models may change. The system plans to add sovereign European solutions from new cloud providers.
Not Just Brussels
Similar incidents are also occurring in Russia. Journalists suspected the use of ChatGPT in drafting bill No. 1126815-8—a mark reading source=chatgpt.com was found in a footnote of the explanatory note. Such "trails" usually remain when copying from the chat. The State Duma explained that a staff member used AI only to quickly find the original source, and the data itself turned out to be correct. However, the committee emphasized that drafting bills requires deep knowledge, which neural networks are not yet capable of handling.
Expert opinion: The European Parliament is demonstrating a pragmatic approach—not to ban, but to control. However, the paradox is that the attempt to reduce dependence on American technology is currently only strengthening it. A real solution could be the active development of Europe's own models, rather than simply redirecting traffic to "verified" services. In any case, the AI solutions market for the public sector is facing a serious transformation.