The Israeli Ministry of Justice published an opinion that aims to clarify whether and how artificial intelligence (AI)-based ventures may use copyrighted content to train machine learning models. According to the Israeli Ministry of Justice, the use of copyrighted content to train a machine-learning tool is likely to be permissible as ‘fair use’ under copyright law and will likely not constitute copyright infringement.

The opinion also explains that in some machine learning projects, Section 22 of the Israeli Copyright Law of 2007, which governs incidental uses of copyrighted works, will apply. Further, if the content used to train the machine-learning tool is deleted at the end of the process, Section 26 of the Copyright Law regarding ephemeral works will apply. The Ministry of Justice believes that these conclusions are consistent with the global trend in the law as applied to machine training.

Yet the Ministry of Justice’s opinion has some caveats. The opinion does not apply to situations in which the copyrighted content used is not diverse, but rather consists of the works of a few individual creators, particularly when that limited content feeds an artificial intelligence tool that generates works to compete with these creators in the markets in which they operate. The opinion also does not discuss the output of machine learning tools. If the output infringes copyrights, the opinion does not create a categorical safe harbor, and a case-specific analysis is required.

The opinion seeks to lift the barrier of legal uncertainty questioning the ability to use copyrighted content to train machine learning tools. It hopes to stimulate desirable activity in the machine learning market, to maximize the competitiveness of Israeli companies.

Click here to read the full opinion (in Hebrew).