<\/p>\n
![\"\"](\"https:\/\/www.europesays.com\/ai\/wp-content\/uploads\/2026\/08\/69d5b2671ef4d.image.png\")
<\/p>\n
Tracy Pratt, Brandfully Yours columnist<\/p>\n
Dear Favorite Business Leader: A year ago, the AI legal storm was still brewing. Lawsuits were filed, arguments staked out, but almost nothing was decided. This summer, several of those cases finally reached real outcomes, and the results are beginning to reshape how businesses need to think about the AI tools they use.<\/p>\n
Here\u2019s what actually landed, what\u2019s still moving and why it matters for your business (with the caveat that a couple of these could shift before publication).<\/p>\n
Price of pirated training data<\/p>\n
The largest of these cases is now closed. On July 20, a federal judge granted final approval<\/a> to Anthropic\u2019s $1.5 billion settlement with a class of authors and publishers in the largest known copyright settlement<\/a> in U.S. history.<\/p>\n It resolves claims that Anthropic had built a research library from more than 7 million pirated books. It\u2019s worth noting that an earlier ruling in the same case still stands, stating that training AI on legally acquired books can qualify as \u201cfair use\u201d (that\u2019s the rule that lets you legally use copyrighted work without permission for things like criticism, teaching and news).<\/p>\n It wasn\u2019t the training itself that cost Anthropic, it was the piracy.<\/a><\/p>\n The implication for business: If you\u2019re using a vendor\u2019s AI tools, it\u2019s worth asking how their models were trained, on what data and whether they\u2019re licensed to use that data.<\/p>\n Human authorship required<\/p>\n In March, the U.S. Supreme Court declined to hear Thaler v. Perlmutter,<\/a> leaving in place a rule that work generated entirely by AI, with no meaningful human input, cannot be copyrighted<\/a>. You might remember this one. It stems from a 2018 attempt to copyright an artwork created by AI.<\/p>\n The implication for business: If your team is using AI to produce code, marketing copy or designs with minimal human editing, that work may not be legally protected, which means a competitor could copy it freely. So, make sure a human is the one shaping and revising anything you want to copyright.<\/p>\n One to watch: NYT v. OpenAI<\/p>\n Of all these cases, this is probably the one worth watching most closely.<\/p>\n In July, the Times and a group of other publishers accused OpenAI of concealing its ability to search training data and chat logs for their copyrighted journalism and asked the court to sanction the company<\/a> over it. That motion was still pending as I wrote this, so a ruling could land before this publishes. The underlying question: Is training a commercial AI model on someone else\u2019s journalism \u201cfair use,\u201d or is it theft?<\/p>\n The implication for business: If the court rules that training commercial AI models on copyrighted journalism is not fair use, then the business model of generative AI providers might pivot toward expensive, permission-based commercial licensing, and that might affect the free chat queries we\u2019ve all enjoyed.<\/p>\n <\/p>\n We\u2019ve graduated from AI chatbots to AI agents<\/a>Trademark, not copyright, may be AI\u2019s sharper exposure<\/p>\n Getty Images\u2019 fight with Stability AI is really two separate cases with two different outcomes so far.<\/p>\n In the U.K., a court largely rejected Getty\u2019s claims in November, though Getty has since won permission to appeal.<\/p>\n In the U.S. case,<\/a> refiled in Northern California, a judge allowed Getty\u2019s trademark and unfair-competition claims to move forward \u2014 not because Stable Diffusion\u2019s AI copied Getty\u2019s photos outright, but because it kept generating images with Getty\u2019s watermark on them, distorted but recognizable.<\/p>\n The implication for business: If you\u2019re running an internal AI tool and it starts producing content with a competitor\u2019s logo baked in, even a garbled version, you may have a trademark problem on your hands before copyright even enters the conversation.<\/p>\n Germany\u2019s warning shot on AI liability<\/p>\n The case I find most interesting is unfolding in Munich.<\/p>\n In late May, a German court issued an injunction<\/a> against Google over its AI Overviews feature, after the AI \u201challucinated\u201d a false story linking a small publisher to a scam and subscription-trap schemes. The court\u2019s reasoning was direct: an AI-generated summary is Google\u2019s own statement, not a neutral list of links, and Google doesn\u2019t get to point to the AI as a separate actor (The Decoder).<\/p>\n Google is appealing, so nothing is final yet. But that same argument \u2014 that an AI provider is the author of its own output \u2014 might eventually force American courts to confront Section 230 of the Communications Decency Act. That\u2019s the law that has long shielded digital platforms from liability for what their users post.<\/p>\n The implication for business: If U.S. judges were to reach a similar conclusion, and I\u2019m not a lawyer, but I\u2019m guessing that might mean a business could potentially be held liable if it runs a chatbot or auto-generated summary that makes a false statement about a person or a competitor.<\/p>\n Also on the radar<\/p>\n Two more worth a mention.<\/p>\n In May, a jury took less than two hours to reject Elon Musk\u2019s $150 billion suit against OpenAI on a technicality \u2014 he\u2019d waited too long to sue. Musk says he\u2019s appealing, so this one isn\u2019t yet fully closed<\/a>.<\/p>\n