Key facts
- The European Parliament is creating an internal AI platform called EPGenAI Hub.
- The platform will offer access to AI models from Meta, OpenAI, Anthropic, and Mistral.
- It aims to provide a safer alternative to public AI tools used for legislative work.
- The platform is currently in testing and could launch as early as September.
- Concerns exist about AI-generated content degrading legislative quality and introducing errors.
The European Parliament is developing an internal platform, EPGenAI Hub, to manage and control the increasing use of artificial intelligence by lawmakers and their staff. This initiative comes amid growing concerns within the institution about the spread of AI-generated parliamentary material, potential inaccuracies, and the degradation of the democratic process. The platform will offer access to AI models from major providers like Meta, OpenAI, Anthropic, and Mistral, operating within Parliament's infrastructure to ensure greater security and data protection compared to public tools.
Around 2,100 individuals, approximately 20 percent of Parliament's staff, reportedly use AI daily for tasks such as drafting speeches, parliamentary questions, and amendments. This widespread adoption has prompted committee chairs to voice alarm, highlighting the risk of "AI slop and misleading AI-generated content" entering legislative work. Lawmakers like Brando Benifei emphasize the need for transparency and leading by example, especially as the EU's AI Act's transparency rules become applicable.
While MEPs are not currently required to disclose their use of AI, parliamentary employees are subject to internal guidelines that mandate labeling AI-assisted work and prohibit the input of sensitive information into public AI systems. The new EPGenAI Hub aims to address these issues by providing a sanctioned, controlled environment. A beta version allows users to select between models hosted securely within Parliament's infrastructure, such as Meta's Llama, OpenAI's GPT-OSS, and Mistral Small, or more powerful external services like OpenAI's ChatGPT and Anthropic's Claude Sonnet.
The rollout of EPGenAI Hub presents a dilemma, as it increases reliance on American AI companies while the EU pursues greater digital sovereignty. This move contrasts with recent decisions, such as replacing Google with the French search engine Qwant and disabling built-in AI features on tablets due to cybersecurity concerns. The Parliament's press service indicated that model offerings may evolve, particularly with the availability of open-source and EU sovereign models from providers contracted by the European Commission.