'AI Model Licenses Masked with Concerning Restrictions'

This week, Google released a family of open AI models, Gemma 3, that quickly garnered praise for their impressive efficiency. However, developers raised concerns about the restrictive license that makes commercial use of the models a risky proposition.

Legal Challenges for Companies

Companies like Meta also present custom, non-standard licensing terms for their openly available models, creating legal challenges for businesses. The uncertainty around the licensing of these so-called “open” AI models is deterring commercial adoption and integration into products or services.

Restrictive Licensing Terms

Meta and Gemma’s licenses have restrictions that limit the ways companies can use the models without fear of legal reprisal. For example, Meta prohibits developers from using the “output or results” of their models to improve other models and imposes user limits for deploying models.

Impact on AI Ecosystem

AI researchers and developers are feeling the impact of these custom licenses, with concerns about enforcement and implications for downstream customers. The fear is that these models may become a trojan horse, limiting their adoption and forcing businesses to stick with less reliable models due to licensing restrictions.

Calls for Collaboration

Despite the success of some models with restrictive licenses, there is a call for providers like Google to move towards more open license frameworks to encourage broader adoption and collaboration with users on accepted terms.

“The lack of consensus on terms and untested assumptions in courts make it all a declaration of intent,” said Jernite. “If certain clauses are interpreted too broadly, good work could end up on shaky legal ground, which is concerning for organizations with successful commercial products.”

Vidal emphasized the urgent need for AI model companies to freely integrate, modify, and share without the fear of sudden license changes or legal ambiguity.

According to Vidal, the current landscape of AI model licensing is confusing, with restrictive terms and misleading claims of openness. Instead of redefining ‘open’ to fit corporate interests, the AI industry should adhere to established open source principles to foster a truly open ecosystem.