Apply AI Summit 2026: Ambition, Trust and Europe’s Next Move in AI
Authored by Dila Şen (Senior Regulatory Compliance Specialist) & Ani Nozadze (Senior Regulatory Compliance Specialist and Senior Team Leader), Adherent
Key Insight
At the Apply AI Summit 2026 in Brussels, Europe’s leaders called for more speed and ambition in AI, alongside AI that is trustworthy, human-centred and sovereign. Regulation is maturing quickly, from the Digital Omnibus to the first harmonised AI standards, and is increasingly seen as a bridge to innovation rather than a barrier. Across the sessions, speakers agreed that trust depends on rigorous evaluation and meaningful human oversight.
Editor’s note: This article reflects the discussions at the Apply AI Summit on 17 November 2026. Later developments may not be covered. Subscribe to our newsletter to stay up to date.
Table of Contents
- Apply AI Strategy, One Year On
- Shaping Europe’s Frontier AI Future: From Ambition to Action
- AI in the Energy Sector for a more resilient energy system
- Deploying Trustworthy AI in Europe: Innovation in the Era of the AI Act
- AI Safety: The Science of Evaluating the Most Advanced AI Models
- Rebooting Automotive Competitiveness Through AI
- Final Reflections
On 17 November 2026, the European Commission brought policymakers, industry leaders, researchers and startups together at The EGG in Brussels, and online, for the Apply AI Summit. The event marked one year since the adoption of the Apply AI Strategy and closed the first European AI Innovation Month. The focus of the Summit was how trustworthy, human-centric AI can boost the competitiveness of Europe’s strategic sectors, from healthcare and mobility to manufacturing and the public sector.
Below, is a summary of the key takeaways from the sessions we followed, in the order they took place.
Apply AI Strategy, One Year On
The day opened with welcoming remarks and a high-level panel that included Roberto Viola, Director-General of DG CNECT at the European Commission, and Jim Hagemann Snabe, the Commission’s Special Envoy for Industrial AI, as speakers, alongside industry leaders.
Roberto Viola presented the EU Blueprint for Industrial AI and framed it as a call to action for Europe to work together. His message to the room was that Europe needs to mobilise, inspire and become very ambitious if it wants AI to deliver for its industries. He also mentioned that three things are necessary for Europe to succeed in becoming a leader in AI – Compute, Capital, and Commitment. The EU is working on all these fronts to succeed, including delivering AI factories, starting to build AI giga factories, start-up and scale-up funds, etc. which were mentioned in EVP Henna Virkunnen’s Keynote as well.
Jim Hagemann Snabe offered a striking historical reference. He recalled the Solvay Conference held in Brussels in October 1927, where many of the greatest physicists of the era met to debate the foundations of quantum theory. Next October will mark its 100th anniversary, and just as that gathering helped shape a century of science, he wants Europe to use AI to create a better future, using Europe’s tradition of responsible technology as an asset, rather than a constraint.
Industry representatives noted a stable environment in capital, meaningful AI adoption and a stable regulatory landscape as some ingredients to success.
Shaping Europe’s Frontier AI Future: From Ambition to Action
The first parallel session in the Auditorium asked how Europe can move from ambition to action on frontier AI.
Building Europe’s research backbone
Martin Brynskov of the University of Copenhagen presented the European Laboratory for Artificial Intelligence (ELAI), which he coordinates. ELAI is a distributed, virtual institute where leading researchers tackle frontier AI challenges such as efficiency, safety and security at European level. Building on a decade of collaboration through the ELLIS network, it spans 17 countries, including Japan and Canada. The European Commission is providing €17 million in seed funding over five years under the “Science for AI” pillar of RAISE, the Resource for AI Science in Europe, and activities begin on 1 January 2027.
Faster, bolder, out of the comfort zone
Nicolas Flores-Herr, Team Lead for Foundation Models and GenAI Systems at Fraunhofer IAIS, was direct: the EU needs to get out of its comfort zone and be faster and bolder.
Trustworthy, human-centred and literate
Catholijn Jonker, Professor of Interactive Intelligence at TU Delft, stressed the importance of trustworthy and human-centred AI. She argued that Europe also needs symbolic AI: unlike large language models, it works with explicit rules and logic, so its reasoning can be traced and audited. Combining the two approaches, known as neuro-symbolic AI, could deliver systems that are both powerful and accountable.
Jonker also called for greater AI literacy, so that people can work with AI systems more effectively. She highlighted the need for meaningful human monitoring: when an AI agent makes a mistake, a human needs to be in a position to catch it.
Sovereignty and the rise of agents
Julien Codorniou, General Partner at 20VC and Board Member of BottleCap AI, a Prague-based developer of efficiency-first large language models, spoke about deploying models in the public sector. He explained that preparing models carefully helps ensure they can be deployed safely for government use.
On the rise of AI agents, the panel agreed that models need to be safe. They also agreed that you can trust most what you have built yourself, which is why Europe’s models need to be sovereign.
AI in the Energy Sector for a more resilient energy system
The parallel session on AI in the Energy Sector focused on trustworthy, interoperable AI in order to create more resilient energy systems in Europe and beyond. Moderated by Merle Becker, the session opened with a keynote from Alejandro Ulzurrun De Asanza Y Munoz, Acting Director for Just Transition, Consumers, Energy Efficiency and Innovation at DG ENER.
Philippe Rambach, Chief AI Officer at Schneider Electric, set out the challenge on several levels.
First of all, data centres themselves need power, and the goal there is to get as many tokens per watt as possible. The grid itself also needs much more AI. But in his view, the most underestimated opportunity is on the demand side. Using AI to optimise heating, cooling and other building systems could save energy. AI can also help in reducing peak demand, which he described as the biggest challenge on the demand side. Global optimisation across the whole system is the ideal, but local optimisation still delivers real value. And underpinning all of it is trust in AI, as without it adoption will stall.
Barbara Wittenberg, CTO of 1KOMMA5°, brought the consumer perspective. Her central message was that flexibility and optimisation depend on data. For that to work, grid data needs to be available and transparent.
Alberto Méndez Rebollo, CEO of Plexigrid, highlighted an economic shift. Thanks largely to renewables, the price of energy itself is falling; however, the cost of the grid is rising. In his view, while Europe might not be able to compete with China or the US on solar panels and large language models, it should invest where it already leads – and energy networks are one of those areas.
The common thread across the panel was that AI’s potential in energy is real, but realising it depends on data access, interoperability and trust. Those are areas where policy and regulation will play a defining role.
Deploying Trustworthy AI in Europe: Innovation in the Era of the AI Act
The afternoon’s AI Act session was moderated by Paul-John Loewenthal, Head of Unit for Regulation and Compliance at the EU AI Office. The panel included Vassilis Karkatzounis of the Hellenic Ministry of Digital Governance, Leonardo Cervera Navas of the European Data Protection Supervisor (EDPS), MEP Brando Benifei, Catherine Vigneron of CEN-CENELEC and Mario Brandenburg of SAP.
The Digital Omnibus and the nudifier ban
The discussion opened with the Digital Omnibus. Agreed by Parliament and Council negotiators on 7 May 2026, it has since been formally adopted as Regulation (EU) 2026/1744 (the Digital Omnibus on AI), published in the Official Journal on 24 July 2026 and in force since 27 July 2026. It pushes back the high-risk rules to 2 December 2027 for high-risk use cases listed in Annex III, and to 2 August 2028 for AI used as a safety component under sectoral legislation. It also adds a new prohibition on AI systems that create non-consensual intimate images of identifiable people, or child sexual abuse material. The ban applies from 2 December 2026.
Panellists stressed how important, and how recent, the ban on non-consensual nudification is. Leonardo Cervera Navas of the EDPS was clear that “move fast and break things” cannot work here. In his view, strong regulation and supervision are a must. He also announced that the EDPS will share the results of its regulatory sandbox in December.
Harmonised standards: the bridge to compliance
Catherine Vigneron of CEN-CENELEC explained how Europe’s AI standards are being built, highlighting cooperation with national standardisation bodies and the wide range of experts involved – from industry, academia and beyond.
She highlighted a key milestone. On 30 October 2025, prEN 18286: Artificial Intelligence – Quality Management System for EU AI Act Regulatory Purposes became the first harmonised standard for AI to enter public enquiry. It has since been published in the EU Official Journal through Commission Implementing Decision (EU) 2026/2274. The standard is designed to help providers of high-risk AI systems meet the quality management requirements of Article 17 of the AI Act, offering a product-focused framework for AI lifecycle governance.
Vigneron explained that companies that comply with a harmonised standard are, in effect, compliant with the regulation. Under the AI Act, applying a standard referenced in the Official Journal gives a presumption of conformity. However, the standards are voluntary, and companies remain free to demonstrate compliance in other ways.
The Commission has asked CEN and CENELEC, working through their Joint Technical Committee JTC 21, to develop standards in ten areas: risk management, data governance and quality, record keeping, transparency, human oversight, accuracy, robustness, cybersecurity, quality management and conformity assessment. According to Vigneron, 14 standards are currently in draft, though quite a few of them are close to the approval phase. She described them as only the beginning, but an essential bridge and tool between the legal text and day-to-day practice.
AI Safety: The Science of Evaluating the Most Advanced AI Models
Moderated by Alessandro Abate, Lead Scientific Advisor at the AI Office, this session brought together Jason Hoelscher-Obermaier, Advisor and former Director of Research at Apart Research, independent researcher Alexandra Olteanu and Chloé Touzet, Policy Lead at SaferAI.
How (not) to test an AI model
In a talk titled “How (not) to test an AI model”, Hoelscher-Obermaier opened with a familiar sight: a product boasting “94% accuracy” on an internal benchmark. His question to the audience: “Does that score tell you what you need to know?” He proposed four checks to apply to any evaluation:
- Risk: What harm is it testing for?
- Cases: Which situations does it cover?
- System: Which model, setup and data sources were tested?
- Grading: How are errors and harm judged?
He closed with a new challenge for evaluators: models can now tell when they are being tested. His practical advice was that an off-the-shelf score may not cover your critical case. Organisations should test the cases their users depend on, in their users’ language, on the system they actually deploy.
What does a benchmark actually measure?
Alexandra Olteanu took the discussion a step further, asking what we need to know before we can trust any claim based on an evaluation. Claims such as “system A performs better than system B”, “system C has achieved superhuman performance” or “system X is less sycophantic” all make assertions about constructs, the underlying concepts being measured. To understand a claim, we need to know how those constructs are defined.
The answer to this issue, Olteanu argued, is measurement modelling. This separates conceptual debates about what we should be measuring from operational debates about how to measure it. Three questions sit at its core:
- How are the constructs defined?
- Is the way they are put into practice valid?
- Is there enough evidence to support the claimed relationship between constructs?
From evaluation to enforceable response
Chloé Touzet of SaferAI focused on what it takes for evaluation to make a difference. At the Dublin Declaration on Trustworthy, Safe and Sovereign AI on 14 October 2026, 24 Member States committed to strengthening Europe’s capacity to assess and mitigate risks from advanced AI. Yet the text leaves key gaps open: an independent evaluation ecosystem, a route to qualify evaluators, guaranteed model access and an agreed methodology.
This matters because European adopters inherit risk from models they did not build. The frontier model is usually developed outside the EU, and the integrator tests only its own application. The adopting organisation is then answerable for a system it cannot fully assess. Meanwhile, providers often choose the evaluations, interpret the results and decide what is acceptable.
Touzet proposed two solutions. The first was a European instrument for AI reliability research, starting in well-defined domains such as railway signalling. The second was stronger evaluation built on four pillars: EU and national capacity, an agreed methodology, guaranteed pre-deployment access, and pre-agreed consequences, so that a finding triggers an enforceable response.
Rebooting Automotive Competitiveness Through AI
The final session we attended focused on mobility. Moderated by Merle Becker, it featured Thomas Skordas, Deputy Director-General of DG CNECT, Taavi Rõivas of Auve Tech (Former Prime Minister of Estonia), Jevgeni Kabanov of Bolt, Vanessa Picron of FORVIA and Markus Johannes Schupfner of QORIX.
The panel focused on autonomous driving, where China is leading and the United States is following. Vanessa Picron, CTO R&D/Engineering at FORVIA, noted that around 70% of cars in China now have Level 2 driver assistance or above, and that the EU is far from this. Speakers agreed that China’s lead comes from having made the sector a national priority. Europe’s strength, by contrast, lies in safety. But Europe needs to move the safety-critical industry at the speed of an AI business.
Markus Johannes Schupfner, CEO of QORIX, argued that there is not yet a real business for autonomous driving in Europe, as robotaxi services are only running in a handful of cities – Zagreb is currently home to Europe’s first commercial robotaxi service, while Madrid is hosting a trial.
The biggest challenge, panellists agreed, is earning public trust, because autonomous vehicles need passengers. Data also came up repeatedly. It needs to stay in the EU, as data is the heart of sovereignty, but Europe also needs its own software. Speakers flagged a practical hurdle too: autonomous systems rely on continuous integration and frequent updates, and end users will not want to update their car every day.
The session’s key messages were protecting the user, collaborating openly, fixing the business model, building a strong EU ecosystem spanning hardware, software and AI.
Final Reflections
A few themes ran through every session we attended. Europe’s leaders want more speed and ambition, and AI that is trustworthy, human-centred and sovereign. Regulation is maturing quickly, from the Digital Omnibus to the first harmonised standards, and these are increasingly seen as a bridge to innovation rather than a barrier. Above all, trust depends on rigorous evaluation, and meaningful human oversight.
For compliance teams, these themes are already part of daily work. New obligations, delayed deadlines, emerging standards and fresh prohibitions are arriving at speed, and organisations need a reliable way to keep track of them. At Adherent, our AI-native regulatory intelligence is built on the same principles discussed in Brussels: trustworthy AI, backed by expert human oversight, that helps product compliance teams monitor, understand and act on regulatory change across the globe.
As Europe moves from ambition to action on AI, staying ahead of the rules will matter more than ever. Book a demo to see how Adherent can help your organisation turn regulatory change into confident compliance.

See Adherent in Action
Discover how agentic AI is reshaping product compliance for global enterprises.
Authors

Dila Şen
Senior Regulatory Compliance Specialist
Leading global regulatory compliance with expertise in artificial intelligence (AI) and batteries.

Ani Nozadze
Sr. Regulatory Compliance Specialist and Sr. Team Leader
Global regulatory compliance professional with expertise in privacy/personal data protection and emerging digital regulations.
