The Parliament's Recipe Against Foreign AIs: EPGenAI Hub, Based on Foreign AI
The European Parliament has launched internal tests of EPGenAI Hub, an artificial intelligence platform designed to give MEPs and their assistants controlled access to models like Claude, ChatGPT, Llama, and Mistral. The stated goal is to curb the widespread use of public AI tools in legislative work, after several committee chairs raised concerns about the quality of amendments and documents written with the help of external chatbots.
The beta version of the platform separates two categories of models. On one side, those hosted directly on the Parliament's infrastructure, including Meta's Llama, OpenAI's GPT-OSS, and Mistral Small from French Mistral AI. On the other, more powerful services managed by external servers, such as OpenAI's ChatGPT and Anthropic's Claude Sonnet.
The final rollout could arrive as early as September, according to two sources familiar with the plans who spoke anonymously. The project has been illustrated to MEPs and staff in a series of workshops, including one held on Tuesday, July 14, presented as an opportunity to discover practical applications capable of making work "smarter and faster."
The Fear of AI Slop in Legislative Texts
Driving the institution towards an internal solution is the growing concern about the spread of AI-generated content in official documents. In a meeting in early July, several committee chairs reported an increase in the number of MEPs and assistants relying on AI to draft amendments and other legislative texts, according to three people familiar with the discussion.
Brando Benifei, the S&D MEP who led the parliamentary work on the AI Act, did not mince words: "The Parliament has not been immune from AI slop and misleading AI-generated content," he stated, adding that MEPs must lead by example now that the transparency rules of the AI Act are becoming applicable.
About 2,100 people, representing 20% of the staff, use artificial intelligence daily in Parliament activities. Signs of undisclosed chatbot use are various: a spike in amendments requiring translation, a recognizable writing style, and, in some cases, references to non-existent laws generated due to hallucination.
Committee chairs have asked Secretary-General Alessandro Chiocchetti to develop guidelines on the use of AI by MEPs. Chiocchetti responded that the issue may fall under the Constitutional Affairs Committee, tasked with amending the internal regulations to introduce stricter rules. Currently, there is no obligation for parliamentarians to declare whether amendments or questions were produced with the help of artificial intelligence.
Parliamentary staff, on the other hand, respond to separate internal guidelines, dating back to April 2024 and already being rewritten. The current rules require labeling AI-assisted texts with the label "AI-assisted" and prohibit the inclusion of sensitive information in public generative AI systems, warning about the risks of inaccuracies capable of spreading misleading information about the institution.
The openness towards American tech giants contrasts with the parallel digital sovereignty strategy pursued by Brussels. In recent months, Parliament has replaced Google with the French search engine Qwant on internal computers and has disabled integrated AI functions on MEPs’ tablets for cybersecurity and data protection reasons. The institution's press service has nevertheless clarified that the model offering is still under review, especially considering the availability of open-source models and recently appointed European sovereign providers entrusted by the Commission to sovereign cloud providers.