The European Parliament — the institution that wrote much of the AI Act — is preparing to give its own lawmakers and staff a controlled way to use generative AI. According to Politico Europe, it plans to roll out an internal platform called EPGenAI Hub as early as September 2026.
A multi-model gateway
The Hub would offer access to models from four providers: Meta (Llama), OpenAI (its open GPT-OSS models and ChatGPT), Mistral (Mistral Small) and Anthropic (Claude Sonnet). Rather than a single vendor, the Parliament is assembling a mix — a notable choice for an institution that will also have to live under the rules it set for AI.
Keeping data in-house
The platform is designed to run on the Parliament's own infrastructure plus external hosting, explicitly to keep sensitive institutional data off public consumer chatbots. Roughly 2,100 staff are in scope, and about 20% already use AI tools — often, the concern goes, by pasting internal material into services the institution does not control. The Hub is partly a governance move: channel that usage into a sanctioned, monitored environment.
Curbing the 'slop'
Part of the rationale is quality control. Parliament has "not been immune to AI slop," and an official platform lets it steer staff toward vetted models and away from unreliable free tools. The irony is real but limited: while it includes US frontier providers, the lineup also features European champion Mistral and open-weight options like Llama and GPT-OSS — not a straight embrace of American closed models.
Practice meets policy
The September timing is "as early as," not a firm launch, and the plan is sourced to reporting rather than an official announcement. But it captures a moment every large organization now faces: the body that regulates AI has to decide, in practice, which models its own people may use — and how to govern them.