Adobe and its piracy "habit": Class action over AI trained on stolen books.
Adobe has been sued: its SlimLM model was allegedly trained on thousands of pirated books. Is the promise of "ethical" AI that is safe for business collapsing? We examine the risks for companies that rely on unverified datasets.
For the past two years, Adobe has been repeating a mantra to us: "Our AI is different. It’s ethical. It’s safe for business." While Midjourney and others ruthlessly scraped the web, Adobe Firefly boasted that it was trained only on licensed stock images. All very nice. Except that, according to a newly filed class action, someone in San Jose "forgot" to check what was inside the text datasets. The result? Adobe has been sued for using pirated books to train its models.
The Matryoshka doll of dirty data
The lawsuit, led by author Elizabeth Lyon, points the finger at SlimLM, a "small" language model optimized by Adobe to run on mobile devices and help manage documents. The problem is not so much Adobe’s code as the "diet" it was fed. Here is the chain of events described in the complaint:
Adobe trained SlimLM on an open-source dataset called SlimPajama-627B.
SlimPajama is a "cleaned-up" version of another dataset, RedPajama.
RedPajama contains "Books3", a notorious collection of 191,000 copyrighted books copied and distributed illegally.
In practice? Without asking anyone for permission, Ms. Lyon’s nonfiction guides (and those of thousands of other authors) ended up in Adobe’s AI stomach.
Misery loves company? No.
Adobe is in good company, but it’s not a company to be proud of. The "Books3" dataset has become Silicon Valley’s kryptonite. It’s the same toxic archive that caused legal trouble for Apple (for its Apple Intelligence) and Salesforce, both accused of using RedPajama without consent, credit, or compensation. The trend is clear: the tech industry is hungry for data, and often eats before checking the expiration date (or the source). Just look at Anthropic (the creators of Claude), which in September agreed to pay astronomical sums—the broader context involves billion-dollar settlements—to resolve lawsuits with authors who accused it of using pirated versions of their works.
Graffico’s Output 🔴
Why is this news a wake-up call for businesses? Because Adobe positioned itself in the B2B market as the "safe" alternative. Businesses pay Adobe specifically to avoid copyright legal troubles. If it turns out that even its "curated" models drew on illegal sources (even indirectly, through unverified open-source datasets), its value proposition collapses.
The lesson for anyone running a business and using AI is doubly important:
Beware the data supply chain:
Using an "open-source" model (like SlimPajama) does not exempt you from legal liability if that model contains stolen material.
The reputational risk:
It’s not enough to say, "I didn’t know." If you sell tools to creatives (as Adobe does), you can’t afford to be caught "stealing" from creatives themselves.
The future of enterprise AI will inevitably depend on certified datasets and, very likely, ones paid for at a hefty price. The "everything is free" party is over.
Ready to bring your ideas to life?
Request a free, no-obligation consultation. Let's talk about your project.
Request a consultation
