NORBERT BIEDRZYCKI

The Copyright Battle: Who Owns the Content Created by Artificial Intelligence?

The Copyright Battle: Who Owns the Content Created by Artificial Intelligence?

On 27 December, The New York Times sued OpenAI and Microsoft, alleging that millions of its articles were used without permission to train the models behind ChatGPT, and that the models can reproduce its journalism almost verbatim. It is the most significant of a series of lawsuits filed this year. Getty Images sued Stability AI in January. A group of authors organized by the Authors Guild, including well known novelists, sued OpenAI in September.

At the same time the US Copyright Office and courts have been answering the opposite question. In August a federal court confirmed that a work generated autonomously by AI, without human authorship, cannot be protected by copyright.

So the generative AI industry faces two copyright questions at once. Was it lawful to train on existing works? And can anyone own what the models produce?

Two legitimate positions

It is easy to frame this as creators against technology companies. The reality is more interesting, because both sides have serious arguments.

Creators and publishers argue that their work has economic value, that it was copied at enormous scale without consent or compensation, and that the resulting products may compete directly with them. A model that can summarize the news may reduce the need to read the newspaper that produced it.

Technology companies argue that training is a transformative use, similar to how humans learn by reading, that models do not store copies of works in any simple sense, and that requiring permission for every piece of training data would make the technology practically impossible to build.

Existing copyright law was not designed for this situation. Courts will have to decide, and they may decide differently in different jurisdictions.

Why this is an investment issue

For investors, this is not an abstract legal debate. It touches the core economics of foundation models.

Litigation exposure. Damages claims in the largest cases could be substantial.

Licensing costs. If training requires licenses, data becomes a significant and recurring cost. Some AI companies have already begun signing content deals with publishers.

Competitive dynamics. Companies with access to licensed or proprietary data may gain an advantage over those that relied on scraping.

Customer risk. Enterprises using generative AI want to know they will not be sued for using its outputs.

That last point already shows up in the market. In September Microsoft announced its Copilot Copyright Commitment, promising to defend commercial customers against copyright claims arising from the use of its AI outputs. Adobe has positioned its Firefly models as trained on content it has the rights to use and offers indemnification to enterprise customers. Legal certainty is becoming a product feature.

A question of foundations

Every durable industry has stable property rights at its foundation. Generative AI is being built on a foundation that is still contested. That does not mean the industry will fail. It means that the eventual settlement, whether through courts, legislation or licensing markets, will redistribute value between those who create content and those who build models.

Leadership question: Can AI companies build sustainable businesses if the legal and ethical foundations of their training data remain contested?

Read next

  1. Frontier AI Safety: Should the Most Powerful Models Be Subject to Different Rules?

    Ethical AI series, written in 2026

  2. The Black Box Problem: Can We Govern What We Cannot Explain?

    Ethical AI series, written in 2026

  3. The EU AI Act: Can Regulation Make Artificial Intelligence Trustworthy?

    Ethical AI series, written in 2026