At Rise of AI 2026 in Berlin, a panel moderated by Anna Gründler with Oliver Stuke (Capgemini), Dr. Irakli Beridze (United Nations) and Jörg Bienert argues against the narrative that Europe lags behind in AI, presenting concrete projects instead. Stuke describes Spark, a German government lighthouse project applying legally sound generative AI to permitting processes such as fishing licenses and wind turbines, built with European partners including Mistral, Aleph Alpha and Fraunhofer institutes and named winner of a global prize for best use of generative AI in government. Beridze describes AI Pol, a UN and Interpol toolkit for responsible AI use in law enforcement, already piloted in India, Brazil, Kazakhstan, Nigeria and Oman with a goal of around 50 countries. Bienert presents SOFI, an open-source sovereign European foundation model trained with Deutsche Telekom and Nvidia, planned to launch in mid-2026 and scale toward 300 to 500 billion parameters. The panel discusses trust as a European differentiator, backed by UN research from Oslo, debates regulatory simplification and AI literacy mandates, and names internal fragmentation as Europe's biggest risk over the next 24 months.
Key takeaways
- Spark, a German government AI project for legal permitting, won a global prize for best use of generative AI in government.
- AI Pol, a UN and Interpol toolkit for responsible AI in law enforcement, is piloted in India, Brazil, Kazakhstan, Nigeria and Oman, aiming for around 50 countries.
- Jörg Bienert's SOFI project plans an open-source sovereign European foundation model, launching mid-2026 and scaling toward 300 to 500 billion parameters.
- UN research from Oslo found trustworthy institutions using flawed AI are trusted more by the public than untrustworthy institutions using trustworthy AI.
- Panelists name internal European fragmentation, not lack of technology, as the biggest risk to Europe's AI leadership over the next 24 months.
Detailed notes
Passion projects: Spark, AI Pol and SOFI
Stuke highlights Spark, a German federal ministry lighthouse project he supports that applies generative AI to legal permitting decisions, from fishing licenses to wind turbine construction, built to be fully legally sound and explainable, developed jointly with European partners including Mistral, Aleph Alpha and Fraunhofer institutes, and awarded a global prize for best use of generative AI in government. Beridze describes AI Pol, a toolkit for the responsible use of AI in law enforcement built with Interpol and funded by the European Commission, launched three years ago, tested in 15 countries and now being rolled out to five large countries outside Europe, India, Brazil, Kazakhstan, Nigeria and Oman, with an eventual goal of around 50 countries. Bienert presents SOFI, an open-source sovereign foundation model developed with seven German universities and research institutions, trained with Deutsche Telekom's industrial AI cloud and Nvidia support, with a first model planned for mid-2026, a 120-billion-parameter model to follow, and plans to scale toward 300 to 500 billion parameters.
Building a European open-source ecosystem
Bienert compares the ambition to Linux breaking the lock-in of proprietary Unix systems in the 1990s, arguing that foundation models are becoming the operating system of the AI ecosystem and that sovereignty requires open-source models, not just infrastructure. He describes the Center for Sovereign AI, which has raised more than 100 million euros with partners including Siemens Energy, Audi and Deutsche Telekom, and an upcoming second OSFM conference bringing together European open-source foundation model initiatives. He warns that the deeper sovereignty risk lies not in cloud infrastructure but in the application ecosystems that Microsoft, Google and Amazon are building on top of their models, which Europe needs an open-source alternative for.
Trust as Europe's competitive advantage
Beridze cites a UN project with the Norwegian government and recent Oslo publication on public trust in AI, finding that trustworthy institutions using imperfect AI are trusted more by the public than untrustworthy institutions using trustworthy AI, and argues Europe's democratic institutions give it an advantage here. Stuke argues 'made in Europe' trusted AI could become a brand comparable to 'made in Germany' for cars, especially for critical infrastructure and government use cases, and frames Europe's dense regulation as a potential asset for 'law as code' applications, since AI systems can be built to navigate that complexity for users in areas from insurance to procurement.
What to fix in regulation
Asked for a regulatory wish, Stuke proposes making AI literacy mandatory across schools, universities and job training. Bienert calls for simplifying Europe's more than 111 AI-related regulations, arguing a startup's third or fourth hire should not need to be a lawyer, and suggests specific rules, such as mandatory labeling of AI-generated content as Norway and France already require for photoshopped images, would be more effective than broad frameworks. Beridze cautions that regulation should support solving problems rather than blocking innovation while waiting for perfect rules, but warns that a catastrophic AI incident could trigger a wave of restrictive regulation that sets development back for years.
What Europe needs to fix in the next 24 months
Stuke names fragmentation as Europe's biggest execution gap, with successful public-sector AI use cases remaining isolated islands by industry and country instead of part of a shared system. Beridze notes that about half of the UN's 194 member states still have no AI or digital strategy, though all European countries do, and warns that failing to close this global gap risks large-scale poverty and migration pressure. Bienert calls for pooling resources instead of competing between German states or European countries, arguing the real competition is with US big tech and China, and points to the German AI Association and the Center for Sovereign AI as steps in that direction.
Quotes
Made in Europe is like a sign that comes with trust, and if the applications are used by European institutions, this gives a good sign to the outside world.
Irakli Beridze
European trusted AI can become a brand like made in Germany is, or was, for cars.
Oliver Stuke
The LLMs more or less become like an operating system of the whole AI ecosystem.
Jörg Bienert
Questions answered in this talk
What is the Spark project mentioned at Rise of AI 2026?
According to Oliver Stuke at Rise of AI 2026, Spark is a German federal government lighthouse project applying legally sound generative AI to permitting processes, from fishing licenses to wind turbine construction, built with European partners and awarded a global prize for best use of generative AI in government.
What is AI Pol and where has it been deployed?
According to Irakli Beridze at Rise of AI 2026, AI Pol is a UN and Interpol toolkit for the responsible use of AI in law enforcement, funded by the European Commission and currently being implemented in India, Brazil, Kazakhstan, Nigeria and Oman, with a goal of reaching around 50 countries.
What is Jörg Bienert's SOFI project trying to build?
According to Jörg Bienert at Rise of AI 2026, SOFI is an open-source sovereign European foundation model developed with seven German universities, trained with Deutsche Telekom's industrial AI cloud and Nvidia support, planned to launch a first model in mid-2026 and scale toward 300 to 500 billion parameters.