“The argument that open-source AI is dangerous because bad actors could misuse it is used to justify not sharing research publicly.”