81 Key Findings of the Israeli Ministry of Justice Opinion Examining first the Israeli perspective, which takes a more permissive approach,, the Ministry of Justice opinion emphasizes Israel’s prominent position as a global leader in AI innovation underscores the importance of clarifying the copyright implications of Training practices. Addressing the legal uncertainties in this area has the potential to significantly accelerate innovation and bolster the competitive advantage of Israeli enterprises engaged in both ML development and content creation. Israeli copyright law currently offers no explicit guidance on whether the use of copyrighted works to train AI systems constitutes infringement, creating what the Opinion describes as a significant obstacle to the advancement of the AI SECTOR and hampers the growth of the ML industry—without delivering corresponding benefits to rights holders. Furthermore, the absence of legal clarity complicates rights enforcement and creates a regulatory grey area that affects all stakeholders across the ecosystem. According to the Opinion, and subject to certain exceptions, the use of copyrighted materials for ML training is generally permissible under existing Israeli legal doctrines. Specifically: • Fair use is likely to apply to most ML-related uses, given the transformative nature of the training process and its minimal impact on the original market for the work. • Incidental use may be relevant in cases where copyrighted content is not the central focus but is included as part of a broader dataset. • Transient use doctrines may provide protection where copyrighted materials are not stored or retained following the completion of the training phase. These conclusions are broadly consistent with emerging legal interpretations in other jurisdictions, which increasingly acknowledge the legitimacy of using copyrighted works in the context of AI training—provided certain conditions are met. Nonetheless, the Opinion sets out important limitations. Notably, the proposed safe harbor does not extend to cases where a ML dataset is composed entirely of works by a single author, particularly where the intent is to replicate or compete directly in that author’s market.
RkJQdWJsaXNoZXIy MjgzNzA=