Don Lovett, Fractional CFO & Managing Principal · ProjectBits Consulting · July 2026

~5 minute read

The owners who stay in control as AI copyright litigation reshapes the frontier models are not the ones who chose the winning model — no one can. They are the ones who governed their context, documented their processes, and kept the model upstream of their business logic before the engagement started. This is the story of one of them. The companion piece, The 20-Minute Move, is how you become one.

On July 10, 2026, four publishers and the novelist Scott Turow sued Google, alleging Gemini was trained on millions of their books and articles without permission. Google’s own internal documents, quoted in the complaint, put the potential liability somewhere between $10 billion and $100 billion. The copyright metadata, the plaintiffs say, was stripped from the content before training.

Most small business owners who use AI read that headline, felt a flicker of unease, and moved on. It’s a story about Google and publishers. Not about them.

I want to tell you about one owner for whom it was about her — and who had already answered the question the lawsuit poses, a full year before it was filed.

She never asked "which model is best?"

Call her Maria. She’s a composite of the owners I work with — not one client, but a pattern I’ve seen enough times to describe as a person. She runs a $2M professional-services firm. Eighteen months ago, an AI vendor pitched her the same thing they pitch everyone: get running in thirty days, let the model handle the workflow, don’t get left behind.

Maria didn’t ask which model was best. She asked a different question, and it turned out to be the only one that mattered: "If the intelligence inside this tool changes, what happens to my business?"

The vendor didn’t have an answer — because the vendor had the whole thing backwards, and so does most of the industry. They start with the solution and go looking for a problem to attach it to. "Here’s AI — where can we put it?" Maria worked the other direction. She started with the business problem — our onboarding is slow and inconsistent, exceptions get handled differently depending on who’s on that day — and only then asked what could solve it. Sometimes the answer was a documented checklist. Sometimes a better-defined role. And in a few specific places, the answer was AI. The AI earned its way in by fitting a named problem; it was never the starting point. That single reversal — problem first, solution second, AI only where it actually fits — is what made everything downstream possible.

So Maria made three decisions before she signed anything.

She documented her processes first. How a job gets priced. How a client gets onboarded. How a reconciliation exception gets handled. All the judgment that had lived in her head and her lead bookkeeper’s head — she got it on paper. Not because a consultant told her to, but because she understood something most owners don’t: you cannot govern a process you cannot see, and you cannot hand a process to AI that you cannot describe. A workflow that only exists in someone’s memory isn’t a workflow. It’s a dependency.

She kept the model upstream, not embedded. She used AI to help her design the decision rules — to map the exceptions, draft the policy logic, think through the edge cases. Then a human validated every rule, and the business ran on those documented rules, executed deterministically. The model helped her build the machine. It was not a part inside the machine.

She named where the model was allowed to decide, and where it wasn’t — in writing, before the engagement started. Where it drafts freely. Where it must stop for her review. What it is never permitted to touch, no matter how confident it sounds.

That’s it. Three decisions. She made them the way Apple named its AI tradeoff before the WWDC keynote rather than during it — the clarity came before the reveal, which is exactly what made it land cleanly.

Then July 10 happened

When the Google suit hit — and the earlier ones, and the Meta suit over 267 terabytes of pirated books with the copyright labels stripped off — here’s what changed for Maria’s business:

Nothing.

Not because she picked the "safe" model. She didn’t; nobody can. The frontier models are all being repriced in courtrooms right now, and when a court orders training data destroyed, the model quietly changes underneath everyone using it. It still answers. It answers with the same confidence. But the depth and reliability in the affected domains shift — on a litigation schedule, not your production calendar — and the model never announces what it forgot.

Maria was insulated from all of that for one reason: her critical business logic didn’t live inside the model. It lived in the documented rules she owned, validated, and could explain to a banker or an auditor at 11pm. The model was upstream of her workflows, not embedded in them. So a settlement that reshapes Gemini’s reasoning doesn’t reshape her month-end close. Her clients still get consistent answers. Her audit trail still holds.

The owner down the street who "got running in thirty days" is not insulated. He built his workflows on the live model’s output. He has no map of what the AI is actually doing inside his business, because the logic was never written down — the AI invented it. When his model shifts, he won’t get a notice. He’ll get a client asking a question he can’t answer.

The quiet part

Maria isn’t a technologist. She never learned the difference between one model and another and she didn’t need to. She was a hero of governance, not of technology — an owner who decided, clearly and in advance, that her business would run on her judgment, extended by AI, not replaced by it.

That is the whole game. The owners who will be fine are not the ones who chose the winning model. They’re the ones who did what Maria did: answered the question — whose intelligence, obtained how, and what happens to me if it changes? — before the engagement, not in the wreckage afterward.

You can still be that owner. The lawsuit is a year-one event in a story that’s going to run for a decade. The window to answer in advance is open right now.

The full argument — why this is a debt you owe whether or not anyone named it, and the architecture that pays it down — is in AI Debt: The Tax on Small Business [projectbits.com/insights/ai-debt-tax-on-small-business/].

Part 2: The 20-Minute Move That Makes You the Owner Who Answered First — the exact thing to do this week.

Don Lovett is the founder of ProjectBits Consulting, a fractional CFO and bookkeeping firm serving small businesses, and the developer of the CFO Operating System™.