Israel - Copyright Use for Machine Learning

Overview

The Office of Legal Counsel and Legislative Affairs at the Ministry of Justice published an interpretive legal opinion on 18 December 2022 addressing whether and when the use of copyrighted works for machine‑learning (ML) training is lawful under Israeli copyright law. The opinion — available from the Ministry — analyzes three statutory exceptions of the Copyright Act of 2007 (fair use, incidental use and transient/ephemeral copy) and concludes that, in most ordinary ML scenarios, creating and using datasets that include copyrighted materials for training will be permissible, subject to important limits. It emphasises that the opinion applies to the learning process itself and not automatically to model outputs, and aims to reduce legal uncertainty for Israeli R&D and commercial AI actors while balancing creators’ interests. See the full text: Ministry of Justice — Uses of Copyrighted Materials for Machine Learning (Hebrew, 18 Dec 2022).

Definitions

Key terms used in the opinion include: "machine learning" (automated training of algorithms from data), "training database/dataset" (collections of works used to feed ML models), "fair use" (Section 19 of the Copyright Act permitting certain uses for study/research etc.), "incidental use" (Section 22 permitting uses where a work appears as an incidental element of another work or process) and "transient/ephemeral use" (Section 26 permitting temporary technical copies that have no independent economic value). The opinion distinguishes between the training process (ingestion, copying and internal representation) and downstream outputs (generated text, images, audio) and treats them as legally distinct questions.

Governance and Institutional Framework

The opinion was issued by the Ministry’s Office of Legal Counsel and Legislative Affairs — the body responsible for legal and legislative interpretation and government legal advice. It is non‑binding guidance rather than primary legislation. The Ministry situates its view within Israel’s copyright statutory framework and references comparative approaches (e.g., US fair use jurisprudence, EU TDM exceptions and policy positions in the UK, Japan and Singapore). While only courts can issue authoritative judicial rulings, the opinion is intended to guide government policy, public agencies and private sector actors and to inform judicial reasoning in future disputes. The opinion also intersects with broader national AI policy documents; for example, Israel’s national AI policy discussions reference aligned regulatory and ethical goals and cross‑ministerial coordination. For the Ministry opinion text see official PDF (Hebrew) and related AI policy materials at Israel's AI Policy 2023 (English).

Key Focus Areas

The opinion focuses on several core legal and policy questions: (1) Whether creation and use of datasets containing copyrighted works for ML training infringes exclusive rights; (2) Which statutory exceptions, if any, apply; (3) Whether contractual prohibitions or technical access restrictions can bar lawful uses; (4) Whether the opinion protects model outputs; and (5) Practical limits and high‑risk scenarios. The Ministry concludes that the three doctrines (fair use, incidental use, transient use) will typically permit ML training: fair use as the principal doctrinal basis (characterizing training as study/research and often transformative); incidental use where works function as means to a technological procedure; and transient use where copies are genuinely ephemeral and deleted after training. Important policy caveats include the need to evaluate market effects (fourth factor), the risk that concentrated datasets built to replicate or compete with a single author’s output will not be protected, and the separate legal assessment required for outputs that may reproduce copyrighted expression. The opinion also cautions that circumvention of technological protection or unlawful access to paywalled content remains problematic, and that contractual clauses may be ineffective only when they are unconscionable standard terms; otherwise, lawful access and contract law remain relevant.

Implementation Framework

Although the opinion is not regulatory text, it recommends a practical compliance approach for ML practitioners: adopt a documented fair‑use analysis for each project; ensure lawful access or licensing where required; design datasets to be diverse and not dominated by a single author where competition risks arise; minimize retention of copies and delete training copies once purpose is served to rely on transient‑use protection; keep logs and provenance metadata for dataset sources; implement reasonable security and access controls; and avoid technological circumvention of paywalls or other access restrictions. The Ministry also highlights that separate legal review is required for model outputs, particularly where outputs might reproduce or be substantially similar to protected works or where outputs target the original author’s market.

Monitoring and Evaluation

The opinion encourages monitoring of market effects and case law developments. It recommends that public agencies, the creative industries and ML developers document use patterns, disputed cases and economic impacts to enable future evaluation and, if necessary, legislative refinement. While immediate regulatory monitoring responsibilities are not imposed by the opinion, it signals the Ministry’s intent to track the legal landscape and to coordinate with other government bodies engaged in AI policy. Practitioners are advised to maintain records to support fair‑use determinations and to enable rapid compliance audits if disputes arise.

Penalties, Liability, and Appeals

As interpretive guidance (not statute), the opinion does not itself create penalties. It reiterates that infringing uses remain subject to remedies under the Copyright Act, including civil injunctive relief, statutory and actual damages, and, in willful cases, criminal sanctions available under Israeli law. The opinion clarifies that reliance on statutory exceptions is a defense to infringement claims where the facts support it; where disputes remain, the courts will decide. Affected parties retain statutory and constitutional rights to appeal judicial determinations. Entities relying on the opinion should still apply standard risk management and consider insurance or licensing where infringement risk is material.

Relationship to Other Instruments

The opinion intersects with Israel’s broader AI policy and with contract, data protection and consumer‑protection regimes. It complements rather than supersedes the Copyright Act (2007) and should be read alongside sectoral rules (e.g., privacy laws when datasets include personal data), contract law (service terms and license agreements), and national AI policy documents. The opinion references comparative international instruments (EU TDM exceptions, US fair use jurisprudence) and is consistent with international movement toward legal clarity for text and data mining. It does not amend existing statutes; rather, it provides an official interpretive stance to be considered by courts and practitioners.

International Alignment

The Ministry’s analysis places Israel closer to jurisdictions that use open‑ended fair‑use frameworks (e.g., United States) while noting EU member states have adopted specific TDM exceptions and other jurisdictions (UK, Japan, Singapore) have taken varied approaches. The opinion cites international case law and policy reasoning to justify its interpretation and presents Israel’s stance as consistent with a global trend favoring permissive treatment of research‑oriented ML training, subject to market‑impact safeguards. The document therefore serves both domestic legal clarity and international policy alignment. For comparative reading see analyses published by law firms and commentators and the Ministry’s AI policy paper at Israel's AI Policy 2023.

Implementation Timeline

Sources and References