Can I train an AI model on code from GitHub?

Short answer: MIT code: Needs review. GPL-3.0 code: Needs review.

Each repository's license decides; training is not named in most code licenses.

MIT code: Needs review

License: MIT License (MIT, curated)

• The license grants broad rights but attaches conditions (such as keeping notices) to copies and adaptations. Whether a trained model, or its output, is a copy or adaptation that carries those conditions is legally unsettled, and in some places training may fall under an exception such as fair use. A human reviewer should confirm this use for your case.

Duties: attribution, license_notice

GPL-3.0 code: Needs review

License: GNU General Public License v3.0 (GPL-3.0-only, curated)

• The license grants broad rights but attaches conditions (such as keeping notices) to copies and adaptations. Whether a trained model, or its output, is a copy or adaptation that carries those conditions is legally unsettled, and in some places training may fall under an exception such as fair use. A human reviewer should confirm this use for your case.

Duties: attribution, license_notice, state_changes, share_alike, source_disclosure

Context

• Whether a model trained on code is a derivative work is unsettled; Rightshold marks it for review where the license does not settle it.

Sources

• GitHub: Licensing a repository

Check it yourself (free, no key)

{
  "tool": "check_license",
  "arguments": {
    "license": "MIT",
    "intended_use": {
      "model_training": true
    }
  }
}

Computed from Rightshold's license profiles; not legal advice. Updated 2026-10-19. JSON · Markdown