Back to Insights
IP · AI · 2026

Intellectual Property in the AI Era: Who Owns AI-Generated Works in Israel?

This article provides general information only and does not constitute legal advice. Each situation is unique — consult with a qualified attorney for guidance specific to your circumstances.

Generative AI — ChatGPT, Midjourney, GitHub Copilot — has shifted from academic curiosity to an everyday business tool for millions. But one question remains open and pressing: who is the legal owner of what AI produces? Is the code, image, or text protected by law? And if so, by whom? This article outlines the current state of Israeli law as of 2026 and its practical implications for startups, creators, and technology companies.

1. Types of Intellectual Property Relevant to AI

Israeli IP law recognises four main branches — each relevant to AI in a different way. **Copyright** protects original works: text, images, code, music. The Copyright Act, 2007, grants rights to the "author" — who, under the prevailing interpretation, must be a natural person. Purely AI-generated output, without meaningful human creative contribution, may not qualify for protection at all. **Patents** protect new inventions with industrial applicability. The Patent Law, 1967, defines "inventor" as a person. The US Federal Circuit confirmed in Thaler v. Vidal (2022) that an AI system cannot be named as inventor — a position the Israeli Patent Office adopts consistently. **Trademarks** protect brands and business names. Here the question is less about who created the mark and more about use — can an AI-generated logo serve as a trademark? Yes, subject to ordinary distinctiveness requirements. **Trade Secrets** protect valuable confidential business information. Complex prompts, trained models, and proprietary AI methodologies may all qualify as trade secrets, regardless of ownership questions over the output.

2. Who Owns AI Output? The State of Israeli Law

This is the burning question — and it still lacks a clear legislative answer in Israel. **Current position:** Under the prevailing interpretation of the Israeli Copyright Act, a work requiring a "human author" will not be protected if created without significant human creative contribution. In other words: a one-line prompt that generated an image — unlikely to be protected. A complex instruction set, editing, curation, and arrangement by a human — much stronger claim to protection. **2026 landscape:** Different countries take different approaches. The UK grants special protection to "computer-generated works" — Israel has not enacted an equivalent provision. The EU (EU AI Act) requires transparency disclosures for AI-generated content. **The practical implication:** If you use AI to generate content for your business, you face two risks: (a) you may hold no copyright in what you "created"; (b) you may be infringing someone else's copyright in works the model was trained on.

3. Training on Protected Data: The Biggest Risk

Now to the point that worries most: **do GPT-4, Midjourney, and Copilot infringe copyright?** AI companies trained their models on vast amounts of internet content — books, articles, code, images — much of it copyright-protected. Multiple lawsuits have already been filed in the US: Getty Images v. Stability AI, Authors Guild v. OpenAI, GitHub Copilot Class Action. **In Israel:** No binding judicial guidance on this yet. Israeli courts will likely be influenced by US law and developments in the EU Court of Justice. **The risk for Israeli startups:** If you train an AI model on protected data without authorisation, you are exposed to infringement claims. If you use a third-party AI API, the provider's terms of service determine who bears liability — usually: you. **Practical recommendations:** - Use training data from licensed open sources (Common Crawl with filters, CC-licensed datasets) - Maintain a log of data sources, download dates, and licences - Include a clear clause in your terms of service on responsibility for AI-generated content

4. IP Strategy for AI Startups

**Freedom to Operate (FTO) Search:** Before launching an AI product, conduct an FTO analysis — checking whether your product infringes third-party patents. Google Patents and ILPTO (the Israeli Patent Office) are starting points. An IP attorney will perform deeper analysis. **Employment and contractor agreements:** As detailed in the separate IP ownership guide, a contractor who develops an AI model for you retains ownership by default — unless a written assignment agreement exists. Include an explicit IP assignment clause in every contractor agreement. **What to register:** - Patents — on unique algorithms, methods, and architectures you developed (not on the idea, but on the specific implementation) - Trademarks — on your product name, logo, company name - Copyright — arises automatically, but documenting versions and dates strengthens your position **Prompts as trade secrets:** Complex prompts you have refined over time that generate business value — keep them confidential. Have everyone with access sign an NDA. Consider whether to publish them as part of a product or protect them as a "secret recipe."

5. 2026 Updates: What Has Changed

**EU AI Act — in force from 2025:** The European law regulates AI risks at four levels. If your company sells to the European market — you have new obligations: transparency about AI use, risk assessment, documentation. Israel is not obligated to implement it, but Israeli companies selling to the EU — are. **ISO/IEC 42001:** New standard for responsible AI governance. Not mandatory, but becoming an expectation in contracts with large clients. **Copyright in code:** A US court ruled (Thaler v. Copyright Office, 2023) that an image created entirely by AI without human creative input is not registrable. This position applies to AI-generated code and other generative content. **Israel in 2026:** The Knesset Economics Committee and the Ministry of Justice are developing a national AI policy. Legislative changes in the copyright and AI space are expected. One piece of advice: do not assume what is permitted today will remain permitted. Build workflows that can adapt.

Checklist

  • Document all data sources used for model training and their licences
  • Include explicit IP assignment clauses in all contractor and developer agreements
  • Implement demonstrable human creative contribution to generative AI workflows
  • Conduct an FTO search before launching a new AI product
  • File patents on unique methods and algorithms
  • Protect complex prompts as trade secrets — NDA + restricted access
  • Update terms of service to include clear statements on AI output ownership
  • Check EU AI Act exposure if selling to the European market

Common Pitfalls

  • Assuming your AI output is automatically protected by copyright
  • Using an AI model API without reading the terms of service — you typically bear liability
  • Developing with contractors without written IP assignment agreements
  • Training on data from unclear sources without licence verification
  • Neglecting version documentation and development process records — critical for due diligence and disputes

FAQ

If I write a prompt and AI generates an image — who owns it?

Under current Israeli law, it is uncertain whether anyone holds copyright over an image created without human creative contribution. The more complex your prompt and the more editing and curation you perform, the stronger your position.

Can I get a patent on an AI algorithm in Israel?

Yes — but not on the abstract idea, only on the specific implementation that is industrially replicable. The Israeli Patent Office examines the 'essence of the invention'. Filing with a patent attorney is strongly recommended.

Is using ChatGPT to create website content for my business legal?

Using the API is licensed. But ownership of the content generated depends on OpenAI's terms and local law. As of 2026, OpenAI grants you usage rights to outputs but does not guarantee copyright ownership.

What does the EU AI Act say and how does it affect an Israeli company?

If you sell an AI service to customers in the EU — the EU AI Act applies to you. High-risk systems (such as AI for hiring, credit, security) require risk assessments, transparency, and documentation.

Can Prompt Engineering be patented?

Generally no — a sequence of textual instructions alone does not meet patentability requirements. If the prompt process is integrated with a unique technical architecture, it may qualify.

Want an IP Strategy for Your Business?

We will build a tailored IP plan together — from FTO searches to patent filings.

Schedule a Consultation