European Parliament staff and MPs recently began testing an internal artificial intelligence platform called EPGenAI Hub, which aims to provide them with a safer environment for using generative AI and reduce the risks of processing sensitive data on publicly available commercial AI tools. This platform will integrate a variety of open weighting models as well as some externally hosted models to meet the already widespread needs for AI use within parliament, while strengthening controls in terms of data security and compliance.
According to Politico, the EPGenAI Hub is currently in the testing phase and is expected to be fully open to Members of the European Parliament (MEPs) and staff as soon as September this year. The platform has been introduced to MPs and parliamentary staff through a number of workshops, one of which was held just last Tuesday. In an email to potential users, the council emphasized that the new platform could help them "work smarter and more efficiently."
In the current beta interface, users can choose between different types of models: an open-weighted model deployed on the European Parliament’s own infrastructure, or a more powerful model provided externally and run remotely. Open weight models include Meta’s Llama, GPT-OSS launched by OpenAI, and Mistral Small from France. In terms of external hosting models, the platform currently supports OpenAI’s ChatGPT and Anthropic’s Claude Sonnet.
It is worth noting that, except for the Mistral Small, which is developed in France, many other models are provided by American companies. In the context that the EU continues to restrain American technology giants through various regulatory measures and attempts to create more European alternatives, this internal platform still relies heavily on American technology, which is considered ironic by public opinion.
The European Parliament Press Office told Politico that the model supply of the EPGenAI Hub is currently being adjusted, especially considering that the European Commission has recently signed contracts with multiple "sovereign cloud" service providers to obtain more open source and "EU sovereign" models. These changes mean that the platform may introduce more AI models developed by European or other non-U.S. entities in the future that are more in line with EU policy orientations at the compliance and sovereignty levels.
Earlier, the European Parliament had adopted a series of measures to reduce dependence on U.S. technology products and services and strengthen data protection. For example, Parliament once changed the default search engine on internal computers from Google to French search company Qwant. In addition, due to network security and data protection considerations, the Parliament has also turned off various AI functions on MPs’ tablet devices. The EPGenAI Hub currently being tested is seen as a concrete practice under this policy line to use “more and more controllable AI” to respond to “new risks brought by AI”.