All hail the AI Eurocrat: Commission rolls out its own ChatGPT-like tool

The European Commission is known more for regulating artificial intelligence tools than deploying them — until now.

The Commission’s IT department, DG DIGIT, has rolled out an AI pilot project to assist staff in drafting policy documents.

While the Commission has been a global front-runner in regulating AI models underpinning tools such as OpenAI’s ChatGPT, it also wants to use AI’s benefits in the European Union executive daily proceedings.

Some of the Commission’s tasks, such as translating its many press releases and rulebooks, are already heavily reliant on AI.

But “GPT@EC,” as the new tool is called, is one of the first attempts to roll out a generative AI tool that resembles mass-market products like ChatGPT, which functions by having a person enter an instruction as a prompt in a chatbox, after which AI generates the requested text, images or video.

DG DIGIT Deputy Director General Philippe Van Damme said on LinkedIn that the department had launched the tool last week.

He said it could help Commission staff create drafts, summarize documents, brainstorm or generate software code.

A Commission official, granted anonymity because they weren’t authorized to speak publicly, confirmed the tool had been rolled out as a pilot project and would be widened to the rest of the Commission staff “in the next few weeks.”

The tool is based on an earlier developed program called GPT@JRC, tested inside the Commission’s science unit, the Joint Research Centre (JRC). A second Commission official sent screenshots, allowing POLITICO to review the tool’s setup.

Staff can choose from several AI models, including Meta’s open-source large language model Llama and OpenAI’s ChatGPT. They can also attach a document, for example, and request that the tool summarize it, or save specific prompts as a template.

Exposing internal documents or data to a third-party tool could pose security risks. When ChatGPT launched in 2022, companies worldwide raced to ensure they had policies in place for when employees used generative AI tools.

The first Commission official said the new Commission-wide AI tool was hosted on the EU executive’s own data center and was designed to align with “corporate rules and compliance standards.”

The tool is designed to “[keep] our data secure,” Van Damme said on LinkedIn.

The Commission has been preparing to use AI in its inner workings for at least a year, with a key document released in January this year.

At the time, AI-driven translation, summarization and document-classification services were already in place, as was a tool to quickly browse EU rulebooks.

The science unit’s GPT@JRC was then name-checked as an example of where the Commission would start with “exploring and testing generative AI.”