Digital Déjà Vu

July 13th 2025

From BitTorrent to neural networks: Why the fight for creative rights feels hauntingly familiar

 

The Scars of Digital Wars Past

I can’t help but be taken back to the days of fighting online piracy with this latest round of AI copyright litigation. I still have some scars and a lot of respect for the folks that fought the fight over the past 15 years. Copyright matters. It is what creativity and innovation are built on. Abuse it and the industries we all love will stand up. Might take a bit of time, and we don’t always get it right (we all remember the SOPA PIPA debacle), but we will stand up. Just as Disney and Universal has against Midjourney in the US, and Getty has done with its litigation against Stability in the UK.

 

The Promise and the Peril

Like much of the industry I get the amazing possibilities AI can bring. Not just in relation to efficiencies and the bottom line, but to expanding the creative world and making content even better for consumers. It is often forgotten that the film industry has always embraced tech whether we consider Warner Bros. transition from silent films to talkies with the Vitaphone system, Disney’s multiplane camera granting visual depth to the animation in Snow White and the Seven Dwarves or George Lucas’s Industrial Light and Magic driving incredible special effects in Star Wars which hold up almost fifty years later. The CEO of Netflix recently said: “There’s An Even Bigger Opportunity To Make Movies 10% Better”. But at the same time AI-led content can’t advance by railroading through centuries of creativity without a penny being paid, without permission, and without any credit.

Nevermind ChatGPT is eating the human world – AI will end up eating itself if we continue down this pathway.

Online video and music piracy was theft on an industrial scale. It was an existential threat to creativity, and during my years fighting it, a threat to the success of companies like Netflix. When content creators, artists and distributors come together to fight infringement, all while creating legitimate alternatives to what is being offered unlawfully, everyone can benefit including the consumer. But it seems like you have to bring legal complaints to agree deals and move legitimate options forward. Can you have one without the other? It seems you can’t. It’s the old carrot and the stick metaphor at work again—pressure the AI companies to do the right thing through legal action while also trying to strike a fair deal.

 

Pirates Then vs Pirates Now

I gave Midjourney a try. My prompt was: “Darth Vader eating ice cream with Shrek”. Within seconds, a perfect rendering appeared. Great for kids wanting to play around with characters they love. The innovation is astounding and to think how this can advance creativity blows my mind.

But, and it is a big but, is this really any different from piracy 15 years ago?

The technology was different—BitTorrent instead of neural networks—but the fundamental position remains unchanged: Someone is getting rich by selling what isn’t theirs to sell.

Back then, pirates operated in the shadows, using pseudonyms and encrypted channels. Over time they found creative ways to profit from the hard work of others (by running ads for well known brands) and at the same time adding legitimacy to what they were doing. Popcorn Time was the threat du jour during my early days at Netflix and it looked pretty legit and not all that different to the Netflix service at the time. It was sometimes hard for a consumer to tell what was legit (meaning licensed or owned) and what was pirate.

Today’s alleged digital buccaneers have venture capital, glossy websites, and ~$300 million in annual revenue. Midjourney doesn’t hide its treasure in digital caves; it displays it proudly on subscription tiers ranging from $10 to $120 per month. And unlike The Pirate Bay its leaders don’t hide in far flung tropical locations. They are sitting pretty in Silicon Valley or Notting Hill.

 

The Scale of Modern Digital Theft

Even with the analogies to the piracy landscape the volume of data involved has also increased and that blows my mind. The Kadrey et al vs. Meta case references data on an almost incomprehensible scale. Meta allegedly torrented “at least 81.7 terabytes of data across multiple shadow libraries through the site’s Anna’s Archive, including at least 35.7 terabytes of data from Z-Library and LibGen” to train its Llama AI models. “Piracy is piracy” – as the GC of Disney recently said. And yet despite the plethora of research on the harms it causes to the industry we still need to convince today’s judiciary (and legislature) of that fact.

The unsealed court documents in the Kadrey case read like a confession from the golden age of BitTorrent. A 2022 internal message from a Meta employee reads, “Using pirated material should be beyond our ethical threshold.” Yet the company allegedly continued anyway, with employees discussing slow download speeds due to limited seeders—the same complaints heard in college dorms twenty years ago.

This case exposes the uncomfortable truth that AI companies are building on the same pirated libraries that we spent decades trying to shut down.

 

When Pirates Don’t Even Pretend to Hide

Trouble is some of our original piracy enforcement playbook didn’t work leaving companies no choice but to file legal complaints. Midjourney in particular behaved like an early digital pirate and simply didn’t reply. Maybe I shouldn’t have been surprised when Midjourney’s CEO David Holz described his service as “kind of like a search engine” and admitted to building his database through “a big scrape of the Internet”. And when asked about artist consent, he responded that “there isn’t really a way to get a hundred million images and know where they’re coming from.” Isn’t it the digital equivalent of claiming you found a truck full of DVDs on the side of the road?

 

Evolution of Theft: From Distribution to Generation

What makes these legal cases fascinating isn’t just the technology—it’s how piracy has evolved from distribution to generation. Traditional pirates simply moved existing content from point A to point B. Midjourney creates entirely new content that happens to look suspiciously like copyrighted characters, raising the philosophical question: Is it theft if the stolen goods are recreated rather than copied? As the case winds through the courts the stakes feel both familiar and yet revolutionary. This is the first time major Hollywood Studios have joined the fight against AI, but I don’t think that it will be the last.

Interestingly the Disney/Universal – Midjourney lawsuit avoids mentioning the Digital Millennium Copyright Act—the very framework we all used for years to send gazillions of notices to pirate sites. It also helped resolve the infamous YouTube litigation and countless other digital disputes. The DMCA’s notice-and-takedown system became the backbone of content protection online, but it assumes a world where infringing content can be identified and removed. What happens when the “content” is an infinite well of potential creations?

 

The New Economics of Digital Piracy

Traditional piracy involved distributing finite copies of existing works. AI piracy involves infinite variations of protectable characters—a Darth Vader in a chef’s hat is still Darth Vader, but proving that becomes exponentially more complex in the world of AI.

The economic models have evolved too. Early pirates gave content away for free, disrupting revenue streams. More sophisticated pirates then added advertising and of course there was the link to organised crime. But Midjourney charges premium prices for what it creates, potentially competing directly with creators’ own licensing deals. When someone can generate professional-quality Pixar-style content for $10 a month, why commission official artwork for $10,000? The damages assessment has to be different as a result and we will find out with the trial in the Anthropic case.

 

Building the Future: Creative Solutions for the AI Age

I think we all learned from the early mistakes with Napster and the music industry’s initial response which was scorched earth litigation—sue the technology, sue the users, sue everyone! I don’t think any creator thinks we can, or indeed, should shut the technology down entirely.

Demanding better controls is smart. Midjourney and others already have technological measures in place to prevent violence or nudity. The implied argument: if you can filter out inappropriate content, you can filter out Darth Vader and Shrek. We used the same argument in the early pirate site blocking litigation against ISPs. BT had technology called Cleanfeed in place to block subscribers accessing child porn sites that were identified in the Internet Watch Foundation list.

Of course figuring out a commercial solution where creators are properly compensated and credited is what we are also pushing for. YouTube’s Content ID system came out of the same thinking. Google didn’t just build technology to detect copyrighted material; it built a revenue-sharing ecosystem that turned potential pirates into partners. Today, many content creators make more money from YouTube than they ever could from traditional distribution.

Many are advocating for a proper licensing scheme and there are lots of providers helping AI find licensed training data. But I do think we need to be creative. What about building consumer tools to adapt the characters they love which automatically pays licensing fees to Disney. Or character generation APIs that ensure authentic representations while compensating rights holders.

I think it is smart to embrace the technology and build alternatives that are built on licensed content. Netflix didn’t defeat piracy just by suing pirate sites; we made piracy a less appealing and less convincing option by building the best streaming service in the world (ok so I am a bit biased!)

 

Fair Use?

This note would not be complete without a reference to fair use (or fair dealing in the UK). In the Kadrey case, Meta is deploying the classic defense playbook: fair use, transformative purpose, and the argument that training AI models on copyrighted material is fundamentally different from distributing pirated content. It’s the same argument Google used in the early days to defend YouTube, and that every major tech platform has deployed when caught using copyrighted material. Today we have two decades of precedent about fair use, transformative works, and the boundaries of technological innovation. From the Sony Betamax doctrine to the recent summary judgment back and forth in the Thomson Reuters vs. Ross litigation, and now the recent summary judgment decisions in the Kadrey and Bartz cases.

It is a fact specific doctrine and so these opinions have not concretely concluded what is fair use in the world of Gen AI. But that hasn’t stopped the Tech companies celebrating their perceived wins. First, came Bartz v Athropic where Judge Alsup said that Anthropic had violated the authors’ rights by saving pirated copies of their books as part of a “central library of all the books in the world” and there will be a trial on damages. But otherwise their training is “exceedingly transformative” and therefore fair use. Second came Judge Chhabria who decided in favour of Meta in agreeing that training is again fair use under US law. During discovery internal Meta messages revealed that the company used pirated datasets with copies of 7.5 million pirated books and 81 million research papers, according to The Atlantic’s LibGen investigation.

Unfortunately and I think surprisingly given the evidence the authors lost the fair use arguments. They weren’t able to show that there was sufficient harm to the market which is part of the US four-part fair use test. He said: “There is no serious question that Meta’s use of the plaintiffs’ books had a ‘further purpose’ and ‘different character’ than the books—that it was highly transformative.”

On the other side of the Atlantic, Getty Images has brought litigation against Stability for alleged infringement of the copyright in their images by Stability’s ‘Stable Diffusion’ model, which generates images from text or image prompts. As of July 2025 the trial is ongoing and the court needs to manage over many issues wrapped up in the case. Of relevance here is Stability’s focus on the analogous fair use principle in the UK of “fair dealing”. It is much narrower than fair use. Stability focuses on the “pastiche” element – namely that an image is pastiche if it is in a style that may imitate another work, artist or period, but is really only inspiration and therefore not copyright infringement. It is not a well decided area in the UK and Stability is capitalising on that.

 

The Real Lesson: Innovation Beats Litigation

I think that the real lesson from the piracy wars isn’t that enforcement doesn’t work—it’s that enforcement alone never works. The companies that thrived were those that anticipated where technology was heading and built business models around inevitable change. The Motion Picture Association calls copyright protection “the backbone of our industry” and supports “a balanced approach to AI that both protects intellectual property and embraces responsible, human-centered innovation.” The fight for creative rights in the age of AI isn’t just about protecting the past—it’s about building a future where technology and creativity can thrive together. The question isn’t whether we can stop AI, but whether we can make it work for everyone.