Open Source AI offers public administrations a way to adopt AI while retaining greater control over the systems, data and infrastructure involved. It can support reuse across administrations, adaptation to local and multilingual needs, interoperability, and reduced dependence on individual suppliers. The importance of Open Source AI is also highlighted in the EU Open Source Strategy, which places open source at the centre of Europe’s technological sovereignty and identifies AI as a critical area for open building blocks, long-term maintenance and public-sector uptake.
On this page, you can find learning resources on what Open Source AI means, why it matters, and the answers to the most common questions on the topic.
Jump to:
Open Source AI definition
Open Source AI refers to AI systems made available under terms that allow anyone to use, study, modify and share them. For machine-learning systems, this requires more than access to source code. The OSI definition points to three core elements:
- Sufficient information about the training data
- The complete code used to train and run the system
- The model parameters, such as weights.
AI openness exists on a spectrum
Following the G7 Vision on AI openness, distinguish:
- Open Source AI with Open Data: code, parameters and full training data are openly available.
- Open Source AI: code and model weights are open, with detailed information on the training data.
- Open Weights AI: model weights are openly available, but data or training code may not be.
- Weights Available AI: weights can be accessed, but use, modification or redistribution may be restricted.
Why it matters for public administrations
Open Source AI can help public administrations:
- understand and inspect how AI systems work
- adapt or fine-tune systems to public-sector needs
- monitor performance, security and reliability over time
- support transparency and accountability
- reduce dependence on a single AI provider
- strengthen reuse, interoperability and digital sovereignty.
EU Policy Context
Open Source AI is linked to several EU policy priorities:
- The EU Open Source Strategy places open source at the centre of Europe’s technological sovereignty. It identifies AI as a critical area for open source building blocks, long-term maintenance, security and sustainability.
- The Strategy is part of the broader Tech Sovereignty Package alongside other AI-related initiatives such as the Cloud and AI Development Act.
- The EU AI Act provides exceptions for free and open source AI systems, unless they are placed on the market or put into service as high-risk systems, prohibited systems, or systems subject to Article 50 transparency obligations.
Checklist before procurement or use
Before procuring or using Open Source AI, verify:
- which components are available: code, weights, data information, documentation and evaluation results.
- which licence terms apply.
- whether the system is maintained, secure and documented.
- whether the model has been tested for the intended public-sector use case.
- who is responsible for deployment, monitoring, maintenance and human oversight.
- whether the solution is compliant with relevant national and EU obligations.
Performance depends on the task, language, data quality, model size, evaluation method and deployment setup. For some use cases, a smaller open model adapted to a specific need may be more suitable than a large closed model, especially for the public sector. For others, a closed service may offer stronger out-of-the-box performance or managed infrastructure.
Public administrations should compare options against their own requirements, including accuracy, relevant EU languages, documentation, data protection, compute needs, cost, maintainability and compliance. They should also consider sovereignty. The key advantage of Open Source AI is that it can strengthen European digital sovereignty when administrations choose European open models, run them on controlled or European infrastructure, and retain the ability to inspect, adapt and maintain them over time.
Public administrations can start with dedicated public sector channels such as the EU Open Source Solutions Catalogue to identify reusable open source tools, including AI-enabled solutions.
For AI-specific resources, model repositories such as the Hugging Face Model Hub allow users to find model checkpoints and related resources, while model cards can provide information on intended uses, limitations, training details, datasets and evaluation results. Software repositories such as GitHub can also be useful for finding AI tools, frameworks and implementations.
Finally, the European Open Source AI Index can help users compare open generative AI models and understand differences in their actual level of openness. The index covers models for text, image, code, video and audio, and is useful for distinguishing between models that are genuinely open and models that are only partially open.
It depends on the use case. Running a small or specialised model may be feasible with limited infrastructure, while training, fine-tuning or serving larger models can require GPUs, cloud infrastructure or high-performance computing. Administrations should therefore assess:
- whether they need to train, fine-tune or only run the model
- expected usage volume and response time
- whether the system should run locally, in a sovereign cloud, or through external infrastructure
- data protection, confidentiality and cybersecurity requirements
- energy use, cost and long-term maintenance
- whether smaller, task-specific or multilingual models would be sufficient.
For access to compute, EuroHPC AI Factories offer computing resources and support services for AI development, and EuroHPC supercomputing access is open to public authorities under specific calls.
Open Source AI can support technological sovereignty, but only if administrations retain control over the AI system, the data it uses, and the infrastructure on which it runs. Before adopting an Open Source AI system, administrations should check:
- what information is available about the data used to train, fine-tune or evaluate the model
- whether the system will process personal, confidential or security-sensitive data
- where prompts, input data, outputs, logs and model interactions are stored or processed
- whether data submitted to the system can be reused by a provider for further training
- whether the system can run on infrastructure controlled by the administration, another public body, or a trusted European provider
- whether the administration can switch hosting, maintenance or support providers without losing access to the model, documentation or operational data.
Public administrations need the capacity to:
- assess licences and openness claims
- understand model cards, datasets and data information
- test accuracy, bias, robustness and cybersecurity
- manage compute, deployment and infrastructure choices
- monitor performance after deployment
- ensure human oversight and accountability
- procure external support without losing strategic control.
These capabilities are also relevant to meeting obligations under the EU AI Act. Where public administrations act as providers or deployers of AI systems, Article 4 requires them to take measures to ensure an appropriate level of AI literacy among staff and others operating or using those systems, taking account of their knowledge, experience, training and the context of use.
G7 establishes a shared language of open source AI
Portugal launches open source AI to drive digital sovereignty
Key Takeaways: OSOR Workshop on Open Source AI
EuroLLM: Pioneering European Open Source AI
