The NetBSD Foundation has presented updated guidelines for contributing to the project’s source code tree. The new rules include a provision prohibiting the incorporation of code generated using AI tools based on large language models, such as ChatGPT, GitHub Copilot, and Code Llama, without prior written approval from the Core Team. The reason cited is uncertainty regarding copyright and non-compliance with the NetBSD licensing policy.
Concerns are related to the fact that AI systems are trained on a vast array of information, including code that is copyrighted and distributed under various licenses. When AI systems generate code, these factors are not considered, and the output may potentially be viewed as a derivative work of the code used in training the model and distributed under specific licenses.
When training the model on code with a license requiring attribution, the generated code by AI tools does not formally meet this requirement, which can be seen as a violation of most open licenses, such as GPL, MIT, and Apache. Additionally, issues may arise regarding license compatibility when inserting into projects under permissive licenses, such as BSD, code generated using models trained on copyleft-licensed code.
Source: opennet.ru
