Learn AI /AI Foundations /Open-source AI
Open-source AI
- Intermediate
- AI Foundations
- Policy officers
- Technical readers
- Managers & front offices
Also called: open source, open-source.
Back to the milestone you came from
In five seconds
What it is
AI systems released under licence terms that let others inspect, use, modify and redistribute source code and other materials needed to run them.
In plain words
It means source code and needed system materials are shared under licence rules that let others inspect, run, change and redistribute them.
An everyday example
A city office wants help finding duplicate records in a registry. One team publishes the checking program, a guide to how it works and a sample setup so other offices in Amman, Beirut and Sarajevo can run it themselves and improve it. Another team only offers a website where people can upload records and get a result back, without showing how the tool works inside. The first approach is closer to open-source AI because others can inspect, adapt and share parts of the system, not just use a closed service.
This analogy is simpler than real AI projects. Real open-source AI may involve several separate components with different licences, such as code, model weights and data notes. Also, a duplicate-checking tool is easier to inspect than a large machine-learning model, whose internal behaviour can remain hard to interpret even when files are openly released.
Remember thisOpen-source is a claim about released components and licence rights, not a mark of quality, safety or approval. Check what is actually available. Weights alone make an open-weight release, not necessarily an open-source one.
This term is contested
Yes. In AI, people disagree about whether releasing only model weights, or only some technical components, is enough to call a system open source.
Understand it
Understand it
Why it matters
For non-specialists, this term matters because openness affects procurement, security review, legal risk, vendor dependence, cost and public trust. An open-source AI project may let your organisation inspect parts of the system, host it internally, commission improvements and avoid being locked into one supplier. But the label alone does not answer the most important questions: exactly what is open, under what licence, with what documentation, and with what safeguards. In policy and leadership settings, those details shape accountability and practical control.
What “open-source AI” can include, and what may still be missing
After viewing, the reader should be able to explain that open-source AI is about which parts of an AI system are openly released, not just whether people can use the output.
01
Openly released parts
An open-source AI project may publish several components, such as code, model weights, setup scripts and documentation.
Left column with stacked labelled boxes: Code, Weights, Setup scripts, Documentation, Evaluation tools. Use solid outlines and green check icons.
02
What outsiders can do
When these parts are available under suitable licence terms, others can inspect, run, modify and sometimes redistribute the system.
Under the left column, add four action icons with labels: Inspect, Run, Modify, Redistribute. Connect them upward to the released parts with arrows.
03
What may still be missing
Some projects do not release training data, full data details or the exact training process. That is why definitions are disputed.
Add a dashed box next to the left column labelled Training data and training details, with a mixed icon such as a question mark or partial indicator.
04
Closed alternative
A proprietary AI service may let people submit prompts or data and receive outputs, while keeping the internal system hidden.
Right column with one visible box labelled App or API. Behind it, faded hidden boxes labelled Code, Weights, Training process. Show a lock icon.
05
Key comparison
The important question is not only whether you can use it, but which parts you can inspect and change.
Bottom row comparison bar: left says Access to internals: broader; right says Access to internals: limited. Use concise text, no decorative elements.
Show two side-by-side columns. Left: an openly released AI project with boxes for code, weights, documentation and evaluation tools, each marked available, and a dashed box for training data details marked variable. Right: a closed AI service with one visible box marked API or web app and hidden internal boxes. Read top to bottom, then compare what outsiders can inspect, modify and redistribute.
Hovering or tapping each box reveals a short note: available, partly available or hidden. A toggle switches the left column between 'code only', 'open weights', and 'broader open-source release' to show that openness comes in degrees.
Describe this without the diagram
This comparison diagram has two columns. The left column represents a more open AI project. It shows code, weights, setup scripts, documentation and evaluation tools as available. Beneath them are four actions outsiders may be able to take under the licence: inspect, run, modify and redistribute. A dashed box labelled training data and training details is marked as variable or only partly available. The right column represents a closed AI service. It shows one visible app or API box, while code, weights and training process are hidden behind it with a lock symbol. The lesson is that open-source AI depends on which internal parts are released, not simply on whether the public can use a service.
With reduced motion: A static two-column comparison replaces hover effects and toggles. Each component is labelled directly as available, variable or hidden, with a short legend explaining the categories.
Open-source AI usually means an AI system whose source code and other materials needed to run it are released under licence terms that allow inspection, use, modification and redistribution. Those materials may include training scripts, documentation, model weights and evaluation tools. Model weights are the numeric settings a trained model uses to produce outputs. A weights-only release is open-weight, not enough by itself to establish that the wider system is open source.
What is released depends on the project. Developers may publish code, a trained model, instructions for running it and sometimes the data or dataset details used for training. A dataset is an organised collection of data. Other people can then test, adapt, repair or compare the system, subject to the licence.
Open release can support scrutiny, reuse and collaboration. It can also make harmful misuse easier. It does not guarantee safety, fairness, legality, interpretability or high quality.
The term remains contested in AI, especially when training data or details are unavailable. Check the source code, model materials and licence rights instead of relying on the label. Proprietary systems keep key parts under the provider’s control, while open-weight models release trained weights without necessarily opening the wider system.
How “A public toolkit for fixing duplicate records” maps onto the real thing
| In the story | In the AI system |
|---|---|
| the published checking program and guide | openly released code and documentation |
| other offices adapting it for local use | others modifying and redistributing the system under licence terms |
| the website-only service | a proprietary or limited-access AI service |
| sample setup for running the tool | deployment instructions, scripts or model files needed to use the AI |
A worked example
An information management officer at a regional hub assesses a released language model for tagging public situation reports before a progress report is uploaded to the portal.
Starting from: The release provides source code, model weights, inference and training scripts, configuration files, evaluation tools and documentation under a licence permitting inspection, use, modification and redistribution. It describes the training sources but does not release the full training corpus or curation records.
- The officer confirms that the licence covers both the code and model weights, rather than relying on the publisher's use of the term “open source”.
- Technical colleagues install the release in an approved environment and reproduce its published evaluation using non-sensitive material.
- They inspect and modify the tagging configuration, then test the adapted model on approved public documents in the required languages.
- The officer records that the training process cannot be fully examined or reproduced because the underlying corpus and curation records are unavailable.
Result: The regional hub can inspect, run and adapt the released system, and may redistribute modifications subject to the licence and organisational clearance. The file nevertheless records the limits of the open-source claim: the release is more than weights-only, but important information about training remains unavailable.
In a United Nations workplace
Following an internal audit observation on undocumented software dependencies, a knowledge management officer in Bogotá must update the file before close of business on a trilingual search assistant used for cleared guidance. The officer checks whether a proposed model described as open source can actually be inspected, hosted, modified and redistributed under its licence.
The label affects what the office can verify and maintain. The officer needs an inventory of the released code, model weights, scripts, configuration files, evaluation tools and documentation, together with the licence terms. A downloadable model with no modifiable code may be open-weight rather than open-source.
Before you try thisOpen-source does not mean approved, secure or suitable. The model and its dependencies still require security, licensing, performance and data protection review. Personal, confidential or security-sensitive material must not be entered into any environment that has not been approved for it.
Where you may hear it
- An ICT governance board reviewing whether a model can be hosted internally
- A procurement meeting when a vendor describes a model as open source
- An internal audit follow-up on software licences and model documentation
- An inter-agency working group comparing reusable language tools
Avoid confusion
Avoid confusion
Often confused with
Open-weight model
A model whose trained numerical parameters are publicly released so others can run, inspect or adapt it, even if the training code, data or licence are not fully open.
Proprietary model
An AI model whose owner uses licence terms or technical controls to restrict access to or reuse of key components, such as its weights or code.
Side-by-side
Common misconceptions
Common misconception: If an AI model is open source, it must be safe and trustworthy.
Correction: Open release can improve scrutiny because more people can inspect or test some parts, but it does not guarantee safe behaviour, legal compliance or good performance. A poorly documented or biased system can be open source too.
Common misconception: Open-source AI always means the training data are public.
Correction: Many projects release code or weights without releasing the full training data. This may happen for legal, privacy, copyright, security or practical reasons. That is one reason the term is contested in AI.
Common misconception: If people can use an AI tool for free online, it is open source.
Correction: Free access and open source are different. A free web service may still hide the code, model weights and training method, and may not allow modification or redistribution.
Common misconception: Open-source AI and open-weight AI mean the same thing.
Correction: Open-weight usually means the trained model weights are available. Open-source often suggests broader access to code and related materials as well. In practice, people sometimes blur these terms, so it is better to ask exactly what has been released.
Quick check
Which statement best distinguishes open-source AI from a closed AI service?
The key distinction is access to internal components and the rights granted by the licence. A project can be open without releasing all training data, and free access or better accuracy does not by itself make something open source.
Go deeper
Go deeper
The simple version
It means source code and needed system materials are shared under licence rules that let others inspect, run, change and redistribute them.
The technical version
In technical use, open-source AI refers to systems whose source code and needed components are released under licences permitting inspection, use, modification and redistribution. Components can include model weights, inference and training scripts, configuration files, evaluation tools and documentation. The term is contested because training data, data curation and optimisation details may remain unavailable. A weights-only release is open-weight, even if its publisher calls it open source. The practical questions are which artefacts are available, what rights the licence grants, and whether outsiders can audit, adapt and deploy the released system.
Reviewed external resources
Primary sourceRecommendation on the Ethics of Artificial Intelligence (opens in a new tab)UNESCO
This link gives a policy and ethics framework for judging AI use and governance, and suits leadership and policy readers who need safeguards beyond the openness label.
Intermediatelink checked 2026-07-31
EncyclopaediaOpen source (opens in a new tab)Wikipedia
This link gives a broad background on what “open source” usually means in software and is useful for readers who need the general idea before applying it to AI.
Intermediatelink checked 2026-07-31
ArticleMachine Learning Glossary (opens in a new tab)Google
This link gives short definitions of machine learning terms and suits readers who want plain-language technical context around models, data and training.
Intermediatelink checked 2026-07-31
VideoVideo explanations: Open-source AI (opens in a new tab)YouTube search
This search results page gives multiple video explanations and suits readers who prefer to compare several introductory presentations rather than rely on one source.
search results, not one pagelink checked 2026-07-31
DiscussionCommunity discussion: Open-source AI (opens in a new tab)Reddit search
This search results page gives community discussions and suits readers who want to see how different users debate the term, while remembering that forum comments vary in quality.
search results, not one pagelink checked 2026-07-31
Where this came from
Drafted with a language model through an offline pipeline, checked against this site's schema, and every external link above fetched to see whether it answers. Treat it as a starting point and check anything that matters against the primary sources. How this site is built and what that means.
The one thing to remember
Open-source is a claim about released components and licence rights, not a mark of quality, safety or approval. Check what is actually available. Weights alone make an open-weight release, not necessarily an open-source one.