⚖️ The Legal Problem of "Probabilistic Copies" in AI
In a new study, Mark Lemley and A. Feder Cooper analyze whether storing the weights of generative models constitutes copyright infringement if those weights can reproduce protected content.
🌍 The research creates a theoretical framework for lawsuits, shifting the focus from the generation process to the mere fact of storing weights as a potentially illegal act.
👤 This is a precedent that could hold developers liable for models even if they do not explicitly output protected content.
Source 1: https://www.linkedin.com/posts/marklemley_probabilistic-copies-in-generative-ai-models-activity-7480249481423147008-sh1 Source 2: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=7067878