# What “open source” means for space-science AI

A model can be downloadable without making every ingredient open. Here is how to inspect the code, weights, data, license and evidence separately.

Canonical: https://brightaifuture.com/guides/open-source-space-science-ai
Format: guide
Source publication: Not established
Bright publication: 2026-09-19
Substantive update: None recorded
Evidence and review: Primary-source review completed 2026-09-19; Bright did not reproduce the cited research.

## Start with permissions, not the label

For a space-science model, “open source” is most useful when it names specific things people can inspect and reuse. Ask separately whether the training and fine-tuning code is available, whether the trained weights can be downloaded, whether the license permits modification and redistribution, and whether the training or benchmark data can be obtained under its own stated terms.

The NASA–IBM Lunar Foundation Model is a concrete example. NASA says the model is hosted publicly, the complete codebase is available, and machine-learning-ready datasets and benchmarks were released alongside it. Its model card identifies Apache 2.0 for the model. Those are stronger, checkable facts than a general claim that the project is “open.”

## Open code, open weights and open data are different claims

Source code describes how software works. Weights are the learned numerical parameters produced by training. Data carries its own provenance, licenses and access conditions. Publishing one does not automatically publish the others, and a permissive code license does not erase restrictions attached to imagery or derived datasets.

A model card should also document intended uses, evaluation tasks and known failure modes. That documentation does not independently prove the model will work on a new crater, instrument or region; it gives researchers a testable starting point.

## What openness makes possible

When the necessary artifacts and permissions are genuinely available, another team can reproduce an evaluation, fine-tune the model for a new mapping task, inspect implementation choices, or report a failure. Openness expands who can ask those questions. It is not a certificate of accuracy, safety, scientific novelty or equal access to the computing resources needed to participate.

## Limitations

Bright did not reproduce the lunar model’s training or benchmark results.

Repository availability can change; license and access terms must be checked at the time of reuse.

Public artifacts do not guarantee that every training-data item is redistributable, complete or equally usable in every jurisdiction.

## Provenance and history

{
  "provenance": {
    "publisher": "Bright",
    "role": "Source-backed evergreen explanation; not a new scientific result"
  }
}

## Original sources

- [NASA, IBM Launch AI Foundation Model for Lunar Science](https://science.nasa.gov/science-research/artificial-intelligence-lunar-foundation-model/)
- [NASA–IBM Lunar Foundation Model: model card](https://huggingface.co/nasa-ibm-ai4science/NASA-IBM-Lunar-Foundation-Model/blob/main/README.md)

## Continue exploring

- [A reusable AI model for mapping the Moon](https://brightaifuture.com/discoveries/lunar-foundation-model)