The $1.5 Billion Parking Ticket
Proposing A Revolutionary Relationship with Data
In This Post:
Retroactive Piracy
18th Century Laws for 21st Century Technology
Putting a Price on Human Originality
‘Good Artists Copy, Great Artists Steal’
The Last Reserves of Exclusively Human Knowledge
New Frameworks for Knowledge
The Choices We Still Have
I knew an obnoxiously rich guy once who treated parking tickets the way you and I treat parking meters. To him, parking tickets weren’t penalties; they were simply the admission price to the exclusive club of ‘Parking Wherever The Hell I Want.’
Rather than circle the block for ten minutes looking for a legal spot like the rest of us peasants, he’d just pull up to the curb wherever he pleased - fire hydrant, loading zone, didn’t matter. The tickets came regularly, sometimes reaching $100 or more, but to him they were simply the price of premium parking. In his mind, he wasn’t breaking the law so much as purchasing an expensive convenience.
I think about that guy a lot these days, especially when I see headlines trumpeting Anthropic’s “historic” $1.5 billion copyright settlement as some kind of watershed moment for AI regulation. The largest copyright payout in U.S. history! A triumph for the creative underclass! A clear message to AI companies that they must respect intellectual property!
What a bunch of horseshit.
Let me explain why this settlement is less a penalty and more a retroactive licensing fee - one that Anthropic can easily afford and that fundamentally changes nothing about how AI companies will operate going forward.
Retroactive Piracy
Here’s what actually happened: Anthropic trained Claude on roughly 500,000 pirated books obtained from shadow libraries like Library Genesis. Authors sued. Anthropic settled for $1.5 billion - about $3,000 per book. As part of the settlement, Anthropic agreed to delete the pirated copies.
That last part is particularly rich. Delete the books? Now? After the models have already extracted every pattern, every stylistic quirk, every narrative structure from those texts? It’s like asking someone to delete a recipe after they’ve already memorized it and opened a restaurant. The knowledge has been absorbed, processed, transformed into model weights. The books themselves are now as useful to Anthropic as an empty eggshell is to an omelet.
But let’s talk about that $1.5 billion figure, because context matters enormously here. At the start of 2025, Anthropic was generating $1 billion in annualized revenue. By August - just eight months later - that figure had quintupled to $5 billion. Current projections suggest they’ll hit $9 billion by year’s end. The company is now valued at $183 billion, triple what it was worth in March.
So this “historic” penalty amounts to what - about six weeks of revenue? Maybe less? Would you pay six weeks’ salary to retroactively purchase something that helped you quintuple your income? Of course you would. Anyone would. For a company experiencing exponential growth, this isn’t a deterrent. It’s a rounding error.
18th Century Laws for 21st Century Technology
The fundamental issue isn’t the size of the settlement - it’s that we’re trying to apply copyright frameworks designed for printing presses to large language models. Our entire legal conception of intellectual property assumes discrete, traceable units of information that can be clearly attributed to specific creators. But that’s not how knowledge works, and it’s definitely not how LLMs work.
Look at any piece of writing - including this post you’re reading right now. Where did these ideas come from? I’ve read thousands of books, articles, papers. I’ve had countless conversations, attended lectures, absorbed information from sources I don’t even remember. Every sentence I write is built on this accumulated knowledge, this vast corpus of text I’ve “trained” on throughout my life.
Am I violating copyright every time I write? Of course not. But the only practical difference between me and Claude is scale and speed. I trained on thousands of texts over decades; Claude trained on millions in months.
Consider how information actually flows in our knowledge ecosystem. Take any book about a historical event - say, Hurricane Katrina. The author draws from:
Public records and government data
Other copyrighted books and academic papers (properly cited, of course)
Original interviews and research
Common knowledge that originated from those same copyrighted sources but has since entered the public consciousness
Now, those copyrighted sources the author cited? They drew from many of the same public records, plus their own sources, which drew from other sources, in an endless recursive loop of knowledge building on knowledge. Trying to untangle who “owns” any particular insight or observation is like trying to identify which specific drop of water makes a river.
This is why copyright law has always focused on expression rather than ideas - you can’t copyright the concept of a boy wizard going to magic school, but you can copyright the specific words J.K. Rowling used to tell that story. It’s why plagiarism requires verbatim copying, not conceptual similarity or the vague sense that you and I both think gravity makes things fall downward.
But here’s the thing about LLMs: they don’t store or reproduce text verbatim. They learn patterns, relationships, statistical regularities. When Claude generates text about Hurricane Katrina, it’s not quoting from any specific book - it’s synthesizing patterns learned from millions of sources into something ‘new.’ Is it original? Or creative? That’s a philosophical question, but to my eyes, the process looks not dissimilar from the process human writers engage in. It’s just global, and a million times faster.
Putting a Price on Human Originality
The optimists claim this settlement will push AI companies to negotiate proper licensing agreements with publishers and authors upfront. They point to deals already struck between some publishers and AI companies as evidence of this trend.
I see the opposite happening.
Think about it from Anthropic’s perspective. They could have spent years negotiating thousands of individual licensing agreements with publishers and authors, each demanding different terms, different payments, different usage restrictions. The legal fees alone would have been astronomical, and the resulting patchwork of agreements would have been a nightmare to manage.
Or they could do what they did: train on everything, build a product worth $183 billion, then pay a one-time settlement that amounts to a few weeks of revenue.
Which would you choose?
This settlement doesn’t create a deterrent - it creates a price list. Future AI companies now know exactly what it costs to train on copyrighted material without permission: about $3,000 per book, payable only if you get caught and sued. For a product that can generate billions in revenue, that’s not a penalty. It’s a bargain.
‘Good Artists Copy, Great Artists Steal’
Here’s what really unsettles creators, though few will say it out loud: the vast majority of what we consider “original” work builds heavily on existing ideas, patterns, and structures. but the rise of GenAI has forced us to confront just how much of our creative process is pattern recognition and recombination. When a machine can read everything ever written and synthesize it into something that feels new, it highlights an uncomfortable truth: most creative work exists on a spectrum of derivation. Yes, there are genuine moments of breakthrough originality - but they’re rarer than we like to admit. Most people are lucky if they have one or two truly original ideas their whole lives (I count myself in that camp, for the record).
This isn’t meant to devalue human creativity. Even Shakespeare borrowed plots. Picasso famously quipped “good artists copy, great artists steal.” The choices we make about what to steal, how to transform it, and why it matters - these remain fundamentally human. But pretending that every book, every article, every creative work springs forth wholly original from its creator’s mind is a fiction we can no longer maintain in the age of AI.
The Last Reserves of Exclusively Human Knowledge
So what information hasn’t been scraped, processed, and transformed into model weights? More than you might think, but it’s shrinking daily.
Consider what remains under exclusively human control:
The thoughts you’ve scribbled in your Moleskin
The conversation you had with your spouse over dinner last night
The therapy session that helped you work through childhood trauma
The business strategy you discussed in a closed boardroom
The novel you’re writing in a Word document on your air-gapped laptop
But here’s the troubling part: we’re actively feeding much of this remaining private information into AI systems ourselves, all day, every day. Every time you ask ChatGPT to help edit your personal essay, every time you use Claude to workshop your business idea, - you’re creating new training data.
I use AI all day, every day, and squirm when I start to think about what it already knows about me. It knows my pet peeves, my finances, my professional challenges, what illnesses my dog has had, what plants grow in my garden . . . multiply that by millions of users, and you begin to see the strategy: we’re not just the consumers of AI - we’re also its ongoing curriculum.
The companies promise this data won’t be used for training. They promise privacy, security, ethical boundaries. But we’ve heard these promises before from tech companies, haven’t we? And even if current leaders keep their word, what about the next leadership team? The next acquisition? The next “pivot”? A few months ago, millions of people nearly had their DNA sold in a bankruptcy proceeding. Their DNA! Now, why someone would send their DNA to the internet in the first place is a mystery to me. That’s literally giving the machines the instructions to make your replacement.
New Frameworks for Knowledge
Here are three concrete approaches we might consider:
The Knowledge Commons
Treat human knowledge as shared heritage. Like air or international waters, research, data, and creative works (after a shortened copyright term, say 10 years) would flow into a global commons. AI could train freely, but in return would contribute back—e.g., by offering free public access to frontier models. It’s open-source principles applied to knowledge itself.
The Ongoing Royalty System
Instead of one-time settlements, AI firms would pay micro-royalties each time a system drew from identifiable sources. Blockchain could track this, creators would be compensated (or credited), and models could cite their human influences. We already do something similar with music streaming.
The Human Curation Premium
As AI generates more content, human value shifts to judgment and context—knowing what matters, when, and for whom. Every correction (“no, that’s wrong”) is unpaid training. We’re all curators of AI’s future knowledge base, yet we subsidize it rather than share in its value. Something about that asymmetry feels off.
The Choices We Still Have
We still have choices about how we engage with these systems, what we feed them, and what we demand in return.
We can demand transparency about how our interactions with AI are used, stored, and potentially repurposed. We can push for regulations that protect not just copyright but also the more subtle forms of intellectual and emotional labor we contribute every time we interact with these systems. We can make laws for this century and this technology.
Most importantly, we can stop pretending that parking tickets will stop the rich from parking wherever they want. The Anthropic settlement shows us that our current legal frameworks are wholly inadequate for the world we’re building. Instead of celebrating pyrrhic victories, we need to imagine and implement systems that recognize the fundamental shift in how knowledge is created, shared, and valued in the age of AI, the way our ancestors eventually recognized that perhaps drilling holes in people’s skulls wasn’t the best way to treat headaches.
The real question isn’t whether AI companies should pay for training data. They should.
But they didn’t, and now it’s mostly moot. The horse has left the barn, learned to drive, and is headed for the highway. The question now is in how we maintain human agency, creativity, and purpose in a world where machines will gradually absorb all the knowledge we give them, and can synthesize nearly anything from it.
That’s a parking ticket everyone is going to have to pay, no matter how rich they are.