European Parliament Launches Internal AI Platform With OpenAI, Meta and Anthropic

Full text

The European Parliament’s answer to its AI worries: More AI

The institution is opening its doors to models from OpenAI, Meta, Anthropic and Mistral as it tries to bring lawmakers’ growing use of the technology under control.

By MAX GRIERA

in Brussels

Illustration by Natália Delgado/ POLITICO

As MEPs and their assistants increasingly turn to artificial intelligence to draft speeches, parliamentary questions and even amendments, the European Parliament is seeking to bring that use under control with an officially sanctioned platform of its own.

The internal platform, known as EPGenAI Hub, will give lawmakers and staff access to models from Meta, OpenAI, Anthropic and France’s Mistral, offering what Parliament hopes will be a safer alternative to the public AI tools already being used for legislative work. The new platform is currently being tested and could begin rolling out as early as September, according to two people familiar with the plans, granted anonymity to discuss the internal strategy.

The plans come as concern rises inside the institution over the spread of AI-generated parliamentary material. At a meeting earlier in July, several committee chairs voiced alarm that a growing number of MEPs and their assistants were using AI to draft amendments and support other legislative work, according to three people familiar with the discussion.

“Parliament has not been immune to AI slop and misleading AI-generated content,” said Brando Benifei, the Socialists and Democrats lawmaker who led Parliament’s work on the AI Act, the EU’s landmark law governing the technology. “MEPs must lead by example as the AI Act’s transparency rules become applicable.”

The project captures the dilemma facing Parliament as generative AI becomes embedded in its daily work. Around 2,100 people use AI daily in the Parliament — around 20 percent of all staff.

Its leaders want officials to stop feeding parliamentary material into uncontrolled public platforms, but they are also concerned that the technology is degrading the quality of legislative work and introducing errors or misleading content into the democratic process. Their suspicions that lawmakers and assistants are already relying on AI have been fueled by a surge in amendments requiring translation, telltale drafting styles and occasional hallucinated references to laws that do not exist.

Committee chairs asked Parliament Secretary-General Alessandro Chiocchetti to draw up guidelines for lawmakers’ use of AI. He replied that the matter might instead fall to the Parliament’s constitutional affairs committee, which would need to amend the institution’s rules of procedure to introduce stricter rules.

MEPs are currently under no obligation to disclose whether amendments, questions or other parliamentary work were produced with the help of artificial intelligence.

Irish Renew lawmaker Barry Andrews, who chairs the Parliament’s development committee, said it was already clear that some MEPs were using AI to draft parliamentary questions, speeches and amendments.

“That is their choice and I fully understand given our workload,” Andrews said. “But for me personally, these are core democratic functions which voters elected us to perform.”

Parliamentary employees are subject to separate internal guidelines on using AI, obtained by POLITICO , although the April 2024 rules are already due to be redrafted. The rules require staff to label AI-generated work with an “AI-assisted” disclaimer and prohibit them from entering sensitive information into public generative AI systems.

The guidelines also warn that inaccuracies produced by the technology risk creating or spreading misleading information or material that misrepresents the Parliament.

The new platform

EPGenAI Hub has already been presented to MEPs and staff at workshops, including one on Tuesday that promised to demonstrate “practical, everyday applications that will help you work smarter and faster,” according to an email seen by POLITICO.

A beta version reviewed by POLITICO allows users to choose between models operated securely within Parliament’s own infrastructure and more powerful services hosted externally.

The locally operated options include Meta’s Llama, OpenAI’s GPT-OSS and Mistral Small. Users can also access externally hosted models including OpenAI’s ChatGPT and Anthropic’s Claude Sonnet.

The rollout will also open the institution further to American AI companies, even as Brussels pursues a parallel effort to reduce the bloc’s dependence on U.S. technology.

That sits uneasily with the Parliament’s own drive for greater digital sovereignty. In recent months, the institution has replaced Google with the French search engine Qwant and disabled built-in AI features on MEPs’ tablets over cybersecurity and data-protection concerns.

The Parliament’s press service said that the AI models offered may change in the coming months.

“The model offerings are undergoing revision, especially in light of availability of open-source and EU sovereign models from sovereign cloud providers recently contracted by the European Commission,” the service said.