Ownership of AI-generated content: who owns it

Service page: Digital commerce and AI lawyer: terms, privacy and direct mailing.

AI is not an “author”: what does this mean for your ownership?

You designed a logo in Midjourney, wrote marketing content in ChatGPT, created music in Suno, and now someone else is using what you produced. “It’s ours,” you say. But is that legally correct? The Copyright Act, 5768-2007, protects an original work (section 4), and the accepted approach is that originality requires a creative contribution by a human being. AI, as far as the law is concerned, is not a human being. So output generated fully automatically, without human creative involvement, is not protected. But reality is more complex than it sounds.

Ownership, if there is any, belongs to the person who used the AI, not to the AI itself and not necessarily to the developers of the system. The practical question is: when exactly is the user considered an “author”?

The originality principle: how much human involvement is enough?

The Copyright Act, 5768-2007, requires a work to be “original”, meaning the product of a human creative choice. Protection for a work created with the help of AI will depend on the degree of human involvement in the process.

If a person wrote detailed instructions (a prompt), chose from several versions, and edited and rewrote the output, there is a reasonable chance that the final work is protected. If a person wrote “write an article about law” and received output without any editing, protection is doubtful. Israeli courts have not yet decided the issue directly.

Note: we have no express, verified Israeli case law on this issue. Follow the developments: this field is changing fast.

Training data: who owns what?

AI companies trained their models on data collected from the web, some of it protected by copyright. The question of who owns the training data is a different question from who owns the output.

If a photo of yours was used to train a model without your consent, a potential infringement may arise under the Protection of Privacy Law, 5741-1981, or the Copyright Act, 5768-2007. In the United States, many lawsuits on the subject are under way. In Israel, the Ministry of Justice took the view, in an opinion of 18 December 2022, that machine learning on protected content will usually be fair use, apart from exceptions, such as a database built from the works of individual authors in order to compete with them; the opinion does not bind the court and does not deal with the output.

Human editing: how much is enough to obtain rights?

If you took AI output, removed parts, added ideas of your own, and shaped the structure, you may be considered the “authors” of the final version, at least as regards your original contribution. The Law recognizes a “derivative work”, an original work based substantially on another work (section 16), and a compilation, whose originality lies in its selection and arrangement (section 4(b)).

The practical conclusion: the more editing, change and creative choice you add, the stronger your legal position.

Training data and “fair use”: where is the line drawn?

Section 19 of the Copyright Act, 5768-2007, permits “fair use” for purposes such as private study, research, criticism, review, journalistic reporting, quotation, or instruction and examination by an educational institution. The Ministry of Justice took the view in its 2022 opinion that AI training will usually be fair use, but no Israeli court has yet decided the question.

One of the considerations in the Law is the effect of the use on the value of the work and on its potential market (section 19(b)(4)). If AI replaces the demand for the original artist’s work, the argument for “fair use” is significantly weakened.

What to do in practice? 4 steps to establish ownership of AI content

• Document your prompts: they are evidence of a human creative contribution.
• Edit the output significantly: add, change, design. Every edit strengthens your position.
• Add a copyright notice: “All rights reserved © [name] [year]”. Israel has no registration of copyright, but when a person’s name appears on the work as its author in the usual way, there is a presumption that they are the author (section 64).
• Read the terms of service: every AI tool sets this differently. OpenAI’s terms of use assign to the user its rights in the output, “if any”; other tools do not necessarily do so.

Employee, freelancer or commissioning party: who owns the rights in the output?

• An employee who created the content for the purpose of their work and in the course of it: to the extent that there is copyright in the output, its first owner is the employer, unless agreed otherwise (section 34 of the Copyright Act).
• A freelancer or a studio you commissioned the work from: the first owner is the author, unless the commissioning party and the author agreed otherwise, expressly or by implication (section 35(a)).
• Transfer of the right: a contract to transfer copyright, or to grant an exclusive license in it, requires a written document (section 37(c)). So it is worth setting in the agreement with the author who owns the output, and what applies if it turns out there is no copyright in it at all.