license

**Understanding AI model licenses** Understanding AI model licenses is essential for legal compliance when deploying systems built on pretrained models, as license terms vary significantly in permissions, restrictions, and commercial use rights. Permissive licenses (Apache 2.0, MIT): allow commercial use, modification, and redistribution with minimal restrictions; Apache 2.0 includes patent grant and attribution requirements; MIT is simpler with just attribution. These are safest for commercial deployment. Copyleft licenses (GPL variants): require derivative works to use same license; complex implications for commercial products; may require open-sourcing code that links to GPL components. Model-specific licenses: Llama has Meta's community license (free for most commercial uses under 700M monthly users, requires Meta approval for larger deployments); some models are research-only (prohibit commercial use entirely). Key considerations: training data licenses (may have separate restrictions), fine-tuned model obligations (inherit base model license), and service versus distribution (some licenses differ for API services versus redistributed weights). Due diligence: verify license compatibility across all components (model, data, dependencies), document license compliance, and consult legal counsel for commercial deployments. License misunderstandings can have significant legal and business consequences.