A marketing agency owner runs a quick prompt through an AI tool: "Write me a blog post about kitchen renovation trends, in the style of Architectural Digest." Twenty seconds later, she has 800 words of publish-ready copy. She posts it to her client's site, closes her laptop, and moves on to the next task.
She has no idea she just created a legal liability that could cost her client tens of thousands of dollars.
That scenario isn't hypothetical anymore. Two major court rulings in the past year — a $1.5 billion settlement involving Anthropic and a decisive loss for an AI legal-research startup against Thomson Reuters — have made 2026 the year copyright law caught up with generative AI. And the businesses now exposed to the fallout aren't just the AI companies. It's the small businesses, freelancers, and marketing shops that use their tools every day.
What Actually Happened in Court
For the past three years, "is AI training on copyrighted material legal?" was an open question that companies mostly ignored because nobody had ruled on it yet. That changed in 2026.
Anthropic's $1.5 billion settlement. In one of the largest copyright recoveries in U.S. history, a federal judge approved Anthropic's settlement with a class of authors who alleged the company trained its Claude models on pirated copies of their books. The deal doesn't resolve the underlying legal question of whether training an AI on copyrighted text is "fair use" — that fight continues elsewhere — but it sets a real dollar benchmark for what copyright exposure can cost an AI company, and by extension, everyone downstream who relies on its outputs.
Thomson Reuters v. Ross Intelligence. This case hit closer to home for anyone using AI to generate written content for a business. A federal court ruled that Ross Intelligence's use of Thomson Reuters' proprietary legal headnotes to train a competing AI research tool was not fair use — a decisive win for the copyright holder. The case is now on appeal to the Third Circuit, but the message to the industry was clear: courts are willing to find that scraping and training on someone else's protected work, without a license, is straightforward infringement, not a shielded "transformative" use.
Together, these rulings tell business owners something important: the legal system is no longer treating AI outputs as a gray area exempt from ordinary copyright rules. And critically, when infringement happens, the lawsuits aren't only aimed at the AI vendor. They're increasingly aimed at the businesses that published the infringing content.
Why the Liability Lands on You, Not the AI Company
This is the part most small business owners miss: the AI itself can't be sued, and in most cases, neither can it fully protect you.
Read the terms of service for the AI content tool your business uses. Many of them include language along the lines of: "you own the output, and you agree to indemnify us against any claims arising from your use of it." In plain English, that means if your AI-generated blog post, product description, or ad copy turns out to closely resemble someone else's copyrighted work, you — not the AI vendor — are on the hook for the legal fees and any damages.
This isn't a theoretical edge case. Courts have already seen disputes where a company's AI-generated marketing image bore an unmistakable resemblance to a well-known film's visual style, triggering exactly this kind of claim. Copyright infringement doesn't require intent — a business can infringe accidentally, simply because the AI model reproduced patterns from its training data a little too faithfully.
For a freelancer or small business, a single infringement claim like this can mean:
- Legal defense costs, even if you ultimately win or settle
- Statutory damages, which under U.S. copyright law can run up to $150,000 per work for willful infringement
- Forced content takedowns, disrupting an active marketing campaign or product launch
- Reputational damage with the client or customer whose content was implicated
And general liability insurance often doesn't automatically cover this. "Personal and advertising injury" coverage — the category that typically covers copyright and trademark claims — varies significantly by policy, and many small business owners haven't checked whether AI-related claims are excluded.
Four Practical Steps to Reduce Your Risk
You don't need to stop using AI tools for content. You need to use them the way you'd use any other business input with legal exposure: with a process.
1. Watch your prompts. Prompts that explicitly request mimicry — "in the style of [named artist/publication/brand]," "make it sound like [named author]" — are the single easiest way to invite an infringement claim, because they create a documented request to copy someone's distinctive expression. Keep prompts focused on your own brand voice, audience, and facts rather than another creator's specific style.
2. Review every public-facing output before it ships. Treat AI drafts the way you'd treat a contractor's work: as a first draft that needs a human check, not a finished product. Look specifically for phrases, structures, or imagery that feel oddly specific or familiar — that's often a sign the model is reproducing something close to its training data rather than generating something genuinely original.
3. Choose vendors that offer indemnification. Not all AI tools carry the same risk. Providers that train on licensed or proprietary data, and that contractually commit to defending customers against infringement claims tied to normal use of their tool, meaningfully shift risk away from you. Read the indemnification clause in your vendor's terms before you build a content workflow around their tool — it's usually a few paragraphs that tell you exactly who pays if something goes wrong.
4. Keep records of how content was made. If a claim ever does surface, your best defense is being able to show you acted in good faith: which tool generated the draft, what the prompt was, what edits a human made, and when it was reviewed and published. A simple log — even a spreadsheet or a dated folder of drafts — can be the difference between "this was an honest, defensible process" and "we have no idea what happened."
The Documentation Habit That Protects You Twice
That last point — keeping a clear, dated record of what happened and when — is a habit that pays off in more than one part of your business. It's the same discipline that protects you when a tax authority asks you to substantiate a deduction, when an insurer asks how a claim arose, or when a client disputes an invoice. The businesses that handle these moments smoothly are almost always the ones that already had good records before they needed them, not the ones scrambling to reconstruct history after a lawyer calls.
That's true for content creation logs, and it's just as true for your financial records. If your bookkeeping lives in scattered spreadsheets or a black-box app you can't export cleanly, you're in the same reactive position as a business with no record of how its marketing copy got written — hoping nothing ever asks you to prove your history.
Keep Your Records as Clean as Your Content Process
Whether it's a copyright claim, an audit, or a lender's due diligence request, the businesses that come out ahead are the ones with a clear, verifiable paper trail. Beancount.io brings that same discipline to your finances with plain-text accounting — every transaction is transparent, version-controlled, and fully auditable, with no vendor lock-in and no black box. Get started for free and build the kind of financial record-keeping habit that holds up whenever someone asks you to show your work.