EU Launches €32 Million Challenge to Advance Trustworthy Cognitive AI

The European Innovation Council has launched its 2026 DeepRAP challenge, with an indicative €32 million budget aimed at advancing AI systems capable of stronger reasoning, abstraction and planning. The programme encourages approaches including neuro-symbolic AI and places interoperability, human oversight, transparency, safety and alignment with the EU AI Act alongside technical development.

Europe Is Putting New Money Into AI That Can Reason and Plan

Artificial intelligence has become increasingly capable of generating text, analysing images and processing enormous amounts of information. The next challenge is making these systems better at deciding what to do with that information when problems require several steps of reasoning, changing plans and understanding uncertainty.

The European Union is now directing research funding toward that problem.

Through the European Innovation Council’s 2026 DeepRAP — Deep Reasoning, Abstraction & Planning towards trustworthy Cognitive AI Systems challenge, Europe is funding research into AI architectures designed to improve reasoning, abstraction and planning while keeping human oversight and trustworthiness at the centre.

The challenge carries an indicative budget of €32 million and is part of the EIC Pathfinder Challenges programme under Horizon Europe. The 2026 EIC Pathfinder Challenges have an overall indicative budget of €96 million.

The application deadline for the 2026 Pathfinder Challenges is 28 October 2026.

The Focus Is Moving Beyond Pattern Recognition

Many modern AI systems are exceptionally good at identifying patterns in large datasets. Large models can process language, images and other forms of information at a scale that would be difficult for humans to match.

But the ability to recognise patterns does not automatically translate into reliable multi-step reasoning.

Consider a system operating in a factory, laboratory or autonomous vehicle. It may need to understand what is happening, determine a goal, create a sequence of actions, monitor whether those actions are working and change its plan if conditions shift.

That is the type of capability the DeepRAP programme is targeting.

The EIC describes the challenge around three central cognitive capabilities: reasoning, abstraction and planning. The selected research portfolio is expected to explore different approaches to achieving those capabilities rather than relying on one predefined AI architecture.

Neuro-Symbolic AI Is One Route Being Explored

One of the approaches specifically encouraged by the programme is neuro-symbolic AI.

Neural networks are particularly effective at learning patterns from complex information such as language, images and sensor data. Symbolic AI, by contrast, represents information through explicit structures such as rules, relationships and logical operations.

Combining the two could allow AI systems to use the pattern-recognition strengths of neural models while incorporating more structured reasoning.

However, the EU programme does not require researchers to build neuro-symbolic systems.

The DeepRAP portfolio can include deep learning, reinforcement learning, neuro-symbolic AI and new approaches inspired by disciplines such as neuroscience, biology and physics. The objective is to investigate multiple technological paths toward trustworthy cognitive AI rather than declare one architecture the solution.

AI Systems That Can Change Their Plans

Planning is another major component of the programme.

A conventional AI system may be designed to execute a predetermined sequence of actions. A more advanced cognitive system would need to account for changing circumstances and potentially revise that sequence.

The DeepRAP research agenda therefore includes areas such as hierarchical planning, contingency planning and continual re-planning.

This could allow an AI system to maintain different possible strategies, evaluate what is happening and modify its actions when new information becomes available.

Such capabilities could be important for machines operating outside controlled laboratory conditions, where unexpected events are unavoidable.

The programme is also designed to explore AI that can operate across different time scales, connecting immediate actions with longer-term objectives.

Multimodal Understanding Could Become More Important

Another focus is the ability to work with different types of information.

Future cognitive AI systems may need to combine language, images, structured knowledge, sensor information and other data sources before making a decision.

For example, an industrial system could potentially combine visual information from cameras with machine specifications, maintenance records and operational rules. A scientific system could combine experimental observations with databases, simulations and published research.

The DeepRAP challenge aims to support research capable of dealing with such multimodal information while handling uncertainty and practical computing constraints.

From Laboratories to Industry and Scientific Research

The potential application areas identified by the EIC extend across several sectors.

The programme lists areas including industry, mobility, civil security, scientific discovery, health, cybersecurity, justice and human-robot interaction.

In scientific research, reasoning-oriented AI could eventually assist with tasks involving hypothesis development, experimental planning and interpretation of information from different sources.

In robotics, stronger planning could help machines respond to changing environments rather than following fixed sequences.

In industrial settings, cognitive AI could potentially support systems that need to balance multiple objectives while reacting to real-time information.

These remain research objectives rather than demonstrated outcomes of the programme. The individual projects funded through the challenge will still have to establish whether their proposed approaches work reliably in specific environments.

Trustworthiness Is Part of the Technical Challenge

The EU is not treating AI capability as an isolated engineering problem.

The DeepRAP programme explicitly incorporates ethical, legal and societal considerations into its research portfolio. Areas include fundamental rights, transparency, privacy, safety and fairness.

That matters because systems capable of making more complex decisions may also require stronger mechanisms for understanding and supervising those decisions.

The programme therefore aims to combine advances in cognitive capabilities with research into how different systems can interact safely and transparently.

The EIC also expects selected projects to participate in shared activities such as interoperability standards, benchmark development and joint pilot demonstrations.

The EU Wants Projects to Work Together

Rather than funding completely isolated research projects, DeepRAP is being structured as a portfolio.

The EIC plans common activities around interoperability, benchmark development, common pilots, multi-agent integration and application shaping.

A shared benchmark could help researchers compare different cognitive AI approaches using common tasks and evaluation methods.

The programme also envisages, where feasible, combining outputs from different projects into modular multi-agent AI systems capable of structured interaction, collective reasoning and planning.

This could become important as AI research moves from individual models toward systems composed of multiple specialised components.

DeepRAP Fits Into Europe’s Wider AI Strategy

The cognitive AI programme is part of a broader European effort to develop advanced artificial intelligence while establishing rules around safety and human oversight.

The EIC describes DeepRAP as supporting Europe’s future position in safe and human-centred AI, while also contributing to competitiveness and technological sovereignty. The programme explicitly connects its objectives with the ambitions of the EU AI Act and Europe’s broader approach to artificial intelligence.

That combination is significant.

The European approach is not simply about developing increasingly capable models. It is also about creating AI systems that can operate within technical, legal and societal boundaries.

What This Could Mean for the Next Generation of AI

The DeepRAP challenge reflects a shift in the questions researchers are asking about artificial intelligence.

The focus is no longer only on whether a model can generate an answer or recognise information. Increasing attention is being directed toward whether an AI system can reason through a problem, form and revise plans, connect different types of information and operate reliably when circumstances change.

Neuro-symbolic AI is one possible route toward these capabilities, but the EIC programme deliberately leaves room for competing approaches.

The €32 million challenge therefore represents an investment in research directions rather than a guarantee of a particular technological outcome.

If the projects succeed, their work could contribute to a new generation of AI systems designed not simply to process information, but to reason and plan within complex real-world environments while remaining subject to human oversight.

For Europe, that makes cognitive AI both a technological research priority and part of a wider effort to shape how increasingly capable artificial intelligence is developed and deployed.

Frequently Asked Questions

1. What is the EU’s DeepRAP challenge?
DeepRAP stands for Deep Reasoning, Abstraction & Planning towards trustworthy Cognitive AI Systems. It is an EIC Pathfinder Challenge focused on advancing AI capabilities in reasoning, abstraction and planning.

2. How much funding is allocated to DeepRAP?
The DeepRAP challenge has an indicative budget of €32 million within the 2026 EIC Pathfinder Challenges programme.

3. What is neuro-symbolic AI?
Neuro-symbolic AI combines neural-network-based learning with structured symbolic representations and reasoning. It is one of several approaches being explored through the DeepRAP programme.

4. What capabilities is the EU trying to develop through DeepRAP?
The programme focuses on reasoning, abstraction, planning, multimodal understanding and trustworthy operation in complex real-world environments.

5. When is the deadline for the 2026 DeepRAP challenge?
The 2026 EIC Pathfinder Challenges, including DeepRAP, have an application deadline of 28 October 2026.