Open AI Licenses: Unveiling Hidden Restrictions

This week, Google unveiled Gemma 3, a family of open AI models that have been lauded for their efficiency. However, developers on X have expressed frustration over Gemma 3's license, which poses significant risks for commercial applications. This issue isn't isolated to Gemma 3; companies like Meta also impose custom, non-standard licensing terms on their open models, creating legal hurdles for businesses. Smaller firms, in particular, are concerned that Google and similar companies might suddenly enforce stricter clauses, disrupting their operations.

Nick Vidal, head of community at the Open Source Initiative, highlighted the problem to TechCrunch, stating, "The restrictive and inconsistent licensing of so-called 'open' AI models is creating significant uncertainty, particularly for commercial adoption. While these models are marketed as open, the actual terms impose various legal and practical hurdles that deter businesses from integrating them into their products or services."

Open model developers often opt for proprietary licenses over standard ones like Apache and MIT for specific reasons. For instance, AI startup Cohere has made it clear that it supports scientific, but not commercial, work on its models. However, the licenses for Gemma and Meta's Llama models include restrictions that limit how companies can use them without legal risks.

Meta's Llama 3 license, for example, prohibits developers from using the model's output to improve any model other than Llama 3 or its derivatives. It also bars companies with over 700 million monthly active users from deploying Llama models without a special license. Gemma's license, while less restrictive, allows Google to restrict usage if it believes there's a violation of its prohibited use policy or applicable laws.

These terms extend to models derived from Llama or Gemma, including those trained on synthetic data generated by Gemma. Florian Brand, an AI researcher at Trier University, argues that such licenses "cannot reasonably be called 'open source.'" He told TechCrunch, "Most companies have a set of approved licenses, such as Apache 2.0, so any custom license is a lot of trouble and money. Small companies without legal teams or money for lawyers will stick to models with standard licenses."

Brand also noted that while companies like Google haven't aggressively enforced these terms yet, the mere threat can deter adoption. "These restrictions have an impact on the AI ecosystem—even on AI researchers like me," he said.

Han-Chung Lee, director of machine learning at Moody's, and Eric Tramel, a staff applied scientist at AI startup Gretel, agree that custom licenses like those of Gemma and Llama make the models "not usable" in many commercial scenarios. Tramel raised concerns about the potential for "clawbacks" and the confusion over licensing for model derivatives. He warned that models could be used as a "trojan horse" to gain insight into successful business cases and then leverage legal action to enter those markets.

Despite their restrictive licenses, some models like Llama have seen widespread adoption, being downloaded millions of times and integrated into products by major corporations like Spotify. However, Yacine Jernite, head of machine learning and society at AI startup Hugging Face, believes they could be even more successful with permissive licenses. He urged providers like Google to adopt open license frameworks and collaborate more directly with users on widely accepted terms.

Jernite noted, "Given the lack of consensus on these terms and the fact that many of the underlying assumptions haven't yet been tested in courts, it all serves primarily as a declaration of intent from those actors. If certain clauses are interpreted too broadly, a lot of good work will find itself on uncertain legal ground, which is particularly scary for organizations building successful commercial products."

Vidal emphasized the need for AI models that can be freely integrated, modified, and shared without the fear of sudden license changes or legal ambiguity. He stated, "The current landscape of AI model licensing is riddled with confusion, restrictive terms, and misleading claims of openness. Instead of redefining 'open' to suit corporate interests, the AI industry should align with established open source principles to create a truly open ecosystem."