Building Europe’s reference framework for autonomous AI.

The strategic goal

The ambition is to establish a European reference framework for autonomous AI: a common foundation through which independently developed capabilities become more useful together, and improvements can transfer between systems.

Models, tools and environments connect through shared interfaces. CEREBRA-AI’s research focuses on how the system chooses and coordinates them, learns from their outcomes and qualifies changes before they are used more widely.

Autonomy within limits set by people

People define the task, permissions, resource limits and required approvals. Within that scope, the framework is designed to select capabilities, carry out a plan, respond to failures and ask for help when needed.

Outcomes feed a separate learning and evaluation process. Proposed improvements must be tested before they become eligible for wider use, with a route to restore a previous version.

Three ambitions, one foundation

Ambition	What CEREBRA-AI is working towards
Scientific	Advance reasoning and continual learning; test whether useful improvements transfer between systems.
Technological	Develop an autonomous runtime, shared interfaces and developer tools, with evidence supporting execution and change.
Commercial	Establish a reusable framework that partners can build on through specialist capabilities, sector applications and services.

Research that strengthens the whole system

• Reasoning connects learned patterns with explicit knowledge, rules and checks.
• Continual learning studies adaptation and the retention of earlier capabilities.
• Grounded interaction connects the framework to people, robots, digital twins and workflows.
• Trust research makes provenance, explanations, uncertainty and oversight part of the same system.

Nine pilots. A common foundation.

Manufacturing, healthcare, mobility, energy, agriculture, coastal resilience, community care and knowledge work bring different demands to the research. The nine planned pilots will test reusable capabilities and shared interfaces in these settings.

A gain in one pilot raises a further question: can another system reproduce it, with compatible components and its own operating constraints?

An ambition to earn through evidence

The project will compare complete-task outcomes using strong baselines, matched capabilities and resource budgets. Evaluation will examine recovery, retained skills, human intervention and resource use alongside performance.

The framework is under development. Its reference status will depend on reproducible results, useful adoption and a technical foundation that others can build on.

The project at a glance

Programme	CEREBRA-AI
Funding	€14,997,790 in European Union funding
Duration	48 months · 01/09/2026–31/08/2030
Collaboration	34 participating organisations across 14 countries
Validation	Nine planned pilot use cases
Coordination	Neuralio AI, Greece
Grant agreement	101299046