AI copyright lawsuits: what SMEs must know in 2026

On September 4, 2026, the Seattle Times and Newsday sued OpenAI and Microsoft in a federal court in New York for copyright infringement. This is neither the first nor the last lawsuit of its kind: since 2023, a wave of litigation over generative AI training has been growing on both sides of the Atlantic. For a small or midsize business using ChatGPT, Copilot or Gemini every day, this legal news is not a distant detail: it directly touches the clauses in its software contracts and the reliability of the content it produces.

In brief

• The Seattle Times and Newsday sued OpenAI and Microsoft on September 4, 2026, in the U.S. District Court for the Southern District of New York, over unauthorized use of their articles to train ChatGPT, Copilot and Bing.
• This lawsuit joins the New York Times case (December 2023) and Richner Communications, which groups about 400 U.S. local newspapers in a complaint filed on June 24, 2026.
• In Europe, a Munich court ruled in November 2025 that memorizing song lyrics inside a model's parameters constitutes reproduction; the Court of Justice of the EU has been examining its first preliminary question on the topic since March 2026 (Like Company v. Google).
• In France, the Alliance de la presse d'information générale (Apig) filed a complaint with the competition authority on August 11, 2026, against Google's AI summaries, and the Senate unanimously voted, on April 8, 2026, to reverse the burden of proof in favor of rights holders.
• Indemnification clauses offered by Microsoft, Google Cloud, OpenAI or AWS exist, but remain conditioned on strict use of the vendor's safety filters: they do not cover everything, nor automatically.

The new lawsuit against OpenAI and Microsoft

In their complaint filed on September 4, 2026, in the U.S. District Court for the Southern District of New York, the Seattle Times and Newsday accuse OpenAI and Microsoft of scraping content from their websites, including paywalled articles, to build the training datasets behind ChatGPT, Copilot and Bing's AI features. Both outlets claim these AI systems can reproduce or closely paraphrase their articles, and cite a 47% decline in search referral traffic to midsize publishers, recorded as of December 2025.

Seattle Times CEO Alan Fisco said: "we feel strongly that we must defend our content, which we spend millions of dollars a year to produce, from being used without our consent or compensation." Microsoft responded that it was "surprised" by the legal action and said it was open to discussion; OpenAI maintains that its models train on publicly available data and that its practice is grounded in fair use under U.S. law.

This case extends an already long line of litigation: the New York Times opened the way in December 2023 against the same two defendants, on the same fair-use arguments. More recently, a collective complaint named Richner Communications, filed on June 24, 2026, in the same court, groups nearly 400 local newspaper titles across 33 U.S. states.

And in Europe: a legal battle taking shape

This is not only an American story. In Germany, music rights body GEMA obtained a ruling in Munich on November 11, 2025, that clearly separates two moments: training a model (where legal exceptions can apply) and a model generating output (where those exceptions no longer hold). The court ruled that memorizing song lyrics inside the parameters already constitutes reproduction.

At EU level, the Court of Justice has been examining its first preliminary question on generative AI and copyright since March 2026, in the Like Company v. Google case (C-250/25), argued before the full court on March 10, 2026. Judges must decide whether training a model constitutes reproduction under copyright law, whether text-and-data-mining exceptions apply, and whether an AI-generated summary amounts to communication to the public. The ruling has not yet been issued, but its outcome will directly affect every European business using AI tools: it will set the liability framework for years to come.

In France, the ground is also shifting on the legislative and competition fronts. On April 8, 2026, the Senate unanimously passed a bill that reverses the burden of proof in favor of rights holders: once evidence makes it plausible that an AI system used a protected work, it is up to the operator to prove otherwise. A few months later, on July 22, 2026, Google activated its AI-generated summaries (AI Overviews and AI Mode) in French search results. Apig, which represents nearly 300 French daily newspapers, filed a complaint with the competition authority on August 11, 2026, arguing that these summaries capture traffic and value from press publishers without negotiation or compensation, under neighboring rights (articles L. 218-1 to L. 218-5 of the French intellectual property code). Note that in March 2024, Google had already been fined €250 million in France for failing to honor its commitments on those same neighboring rights.

What this actually changes for an SME

A small or midsize business will almost never be a party to these lawsuits. But it is a customer of the companies that are: OpenAI, Microsoft, Google. Three practical consequences follow.

First, indemnification clauses. Since late 2023, Microsoft, Google Cloud, OpenAI and AWS have offered guarantees meant to cover business customers facing infringement claims tied to their AI's output. In practice, these commitments remain conditional: they often require using the vendor's safety filters exactly as specified, they evolve as new problematic content is discovered, and none of the major vendors has publicly detailed how they would actually work in a real dispute.

Second, the risk of unintended reproduction. A model that has memorized a protected excerpt can, in rare cases, output it almost verbatim. For a business publishing AI-generated content (blog posts, product descriptions, marketing materials), this creates a reputational and legal risk if that content turns out to be a disguised copy of someone else's work.

Third, the regulatory uncertainty itself. As long as the CJEU has not ruled on the Like Company case, and as long as the American lawsuits remain unresolved, the rules of the game can still shift. A prudent business documents its AI usage and keeps watching, rather than treating the topic as settled once and for all.

FAQ

Can my business be sued if AI-generated content reproduces a protected work?

The direct legal risk mostly falls on the model provider (OpenAI, Microsoft, Google), not the end user. But if your company publishes that content under its own name, it takes on editorial responsibility, regardless of the tool used to produce it.

Do AI vendor indemnification clauses really protect my business?

Partially. Microsoft, Google Cloud, OpenAI and AWS have offered guarantees since 2023, but they are conditioned on strict use of their safety filters and have never been publicly detailed for a real dispute. Read the precise terms of your contract, not just the marketing announcement.

What will the CJEU's expected ruling in Like Company v. Google change?

This preliminary question, argued on March 10, 2026, must clarify whether training a model constitutes reproduction under European copyright law. The ruling will set a reference framework for the whole European Union, with a direct effect on the contracts of AI vendors operating in France.

Should businesses stop using ChatGPT, Copilot or Gemini while these lawsuits are pending?

No. No ruling to date bans business customers from using these tools. Best practice is to document usage, have published content reviewed, and follow how contractual clauses evolve, rather than giving up tools that have become mainstream.

To go further on the legal obligations shaping AI use in business, see our guide on the European AI Act or explore how other businesses structure their AI usage in our customer success stories.