Copyright Is Dead. Copyleft Is Too.

February 17, 2026

In 2000, Metallica sued Napster.

If you're old enough, you remember the sides: Lars Ulrich at the Senate hearing, the printout of 335,435 usernames, the righteous fury of artists being "stolen from". It was the defining copyright battle of the internet age. Metallica became the face of enforcement — the line between creators' rights and digital anarchy.

Their 1998 album Garage Inc. includes "Whiskey in the Jar." It's a traditional Irish folk song. It existed around campfires for centuries before anyone thought to record it. Thin Lizzy put it on vinyl in 1972. Metallica put it on a CD, copyrighted their performance, and then sued the platform that let people share music.

They copyrighted their rendition of a song that belonged to no one, won a Grammy for it, and then became the most aggressive enforcers of the enclosure.

In the eighties, we taped songs off the radio and it didn't feel wrong. In the late nineties, Napster felt no more wrong. Because it wasn't. Copyright was what was wrong — but big blue FBI warning labels on your VHS tapes had spent decades spreading the message: we are due our pound of flesh.

The First Funeral: Copyright

Napster didn't kill copyright. The switch to digital media did. Copyright died the first time a copyrighted work became a file — a sequence of bits that could be copied perfectly, infinitely, at zero cost. By the time ARPANET made file transfer trivial in the early 1970s, the body was already cooling. Most just hadn't noticed.

Napster was just the moment everyone noticed. The underlying reality — costless copying and distribution at distance — had already made copyright's economic basis obsolete. Copyright was designed for a world where reproduction was expensive: printing presses, vinyl stamping, physical distribution. In that world, controlling the copy was controlling the revenue. When copying became free, the revenue model died. Everything after was life support.

The life support was aggressive:

In the US: DMCA takedowns, Comcast forging network packets to kill BitTorrent connections, the RIAA suing grandmothers for music their grandchildren downloaded
In Canada: law firms like Aird & Berlis filing claims against thousands of people demanding $750 to $5,000 each — while the Motion Picture Association told Parliament that suing individuals was "not part of their legal strategy"
In France: the government built an entire agency to disconnect pirates from the internet. In four years it managed to disconnect exactly one person, for fifteen days, before the provision was repealed

Each escalation more desperate than the last. None of it worked. The sharing moved to private trackers, invite-only networks, Libgen — underground, not gone.

And then the distribution capture pattern completed its cycle. Napster was killed. BitTorrent was domesticated or scared underground. In their place: Spotify, Netflix, YouTube. Centralized platforms that look like access but function as capture. The distributed system didn't lose on technology — millions of people had coordinated voluntarily, shared resources, built something that worked. It was outlawed, then replaced with a rental model.

The rental model works for the institutions. It does not work for the creators. Spotify pays between $0.003 and $0.005 per stream. An artist needs roughly 350,000 streams to earn a month's minimum wage. The old system paid artists poorly. The new system pays them almost nothing while claiming to have solved the problem.

Copyright died with digital transmission. The enforcement failed. The platforms captured what was left. And now the AI companies have arrived to confirm what it always was.

Anthropic settled a copyright lawsuit for roughly $1.5 billion. That sounds like accountability until you do the math: approximately $3,000 per book, across around 500,000 titles downloaded from piracy sites like LibGen. Three thousand dollars for a novel that someone spent years writing.

As Thea Lim wrote in The Walrus after her own novel turned up in an AI training dataset: the settlements don't protect artists. They legitimize corporate data extraction. Copyright becomes a price tag. And a price isn't a prohibition — it's a cost of doing business. The same way a speeding ticket doesn't prevent speeding. If you're rich enough, it's a line item.

And then the mask slipped entirely. Disney sent ByteDance a cease-and-desist over Seedance 2.0, its new AI video tool, for generating Star Wars and Marvel characters without authorization. Disney's attorney called it a "virtual smash-and-grab." SAG-AFTRA condemned the unauthorized use of actors' likenesses. The MPA demanded ByteDance cease all infringing activity.

Righteous fury. Except: Disney invested $1 billion in OpenAI and licensed over 200 of its characters — the same characters — to Sora for AI-generated video. OpenAI trained on the same copyrighted internet as everyone else. The difference between Sora generating Darth Vader and Seedance generating Darth Vader isn't the technology. It's who signed the check.

Disney isn't protecting its art from AI. It's protecting its revenue share from AI. The cease-and-desist isn't a copyright claim. It's a licensing negotiation conducted through lawyers instead of business development.

Copyright enforcement only ever functioned as prohibition against people who couldn't afford the fine. Against individuals — and individuals were never the real threat to Copyright.

Copyright is dead. What we see in the news and haunting courtrooms, collecting settlements, and calling it justice is the spectre of an idea bastardized to charge rents.

The Second Funeral: Copyleft

In 1980, Richard Stallman couldn't fix a printer. The Xerox 9700 at MIT's AI Lab kept jamming, and Stallman — who had modified the previous printer's software to alert users — couldn't touch the new one. The source code was proprietary. Xerox refused to share it. A researcher at Carnegie Mellon who had access refused too — he'd signed an NDA.

From that frustration, Stallman built the GNU project and the GPL. The founding insight of copyleft: if you can't read the code, you can't fix the printer. So make the code open, and make openness a condition of use.

It worked for forty years. And now an LLM can write a printer driver from the spec sheet without ever seeing Xerox's source code. The founding grievance is obsolete.

Copyleft — the GPL, Creative Commons, share-alike licenses — was the counter-movement. Where copyright said you can't copy this, copyleft said you must share your changes to this. It was built on a foundational assumption: that the code, the text, the artifact is the valuable thing. You license the artifact. You protect the artifact. If someone modifies your GPL'd code, they must release their modifications. The artifact stays open.

That assumption held as long as writing code required skilled human labor. The code was scarce because producing it was expensive. The license had teeth because the artifact had value.

The cracks showed before LLMs. Amazon took Elasticsearch — built in the open under a permissive license — and offered it as a managed cloud service without contributing back. Elastic switched to the SSPL — trying to use copyright to control not the code but access to the running code. Amazon forked it and built OpenSearch. MongoDB did the same. HashiCorp abandoned the MPL for the BSL in 2023 — the community forked Terraform within two weeks. Redis relicensed. Each one a company that built its success on open source, then pulled up the drawbridge because the license couldn't protect them from someone with deeper pockets using their own code against them.

The licensing panic was the immune response. The body was already sick.

Copyleft died the minute a large language model became better at coding than most coders. In my estimation, that was when Claude Sonnet 4.5 was released in September 2025.

If someone reads your GPL'd codebase, writes a specification describing what it does, and has an LLM produce a clean-room reimplementation in twenty minutes — what did the license protect? The text. Not the knowledge. Not the design. Not the intent. The specific arrangement of characters in a file that can now be regenerated from description alone. I generated 34,000 lines of code in four days for $250. No claim it works in production — but the artifact itself is no longer the bottleneck.

You can't GPL a specification. You can't Creative Commons a design pattern. The license attaches to the artifact, and the artifact is no longer where the value lives. The specification is the skill now — and no open framework exists to keep specifications free the way the GPL kept code free.

Not all open-source licensing is dead. MIT and BSD-style permissive licenses survive because they never tried to control the artifact — they just ask for attribution. "Tell people where you got this." That's not enclosure. That's the campfire tradition: "I heard this from so-and-so." Attribution doesn't depend on scarcity. It's what Metallica could have done with Whiskey in the Jar — just say where the song came from. It's not theft to hear a song and sing it. It's not theft to read a book and learn from it. It's not theft to read code and understand the idea behind it. What dies is the claim that the artifact itself can be owned and controlled.

I have argued that knowledge was always free and that enclosing it in artifacts was the anomaly. Copyleft tried to keep knowledge open by protecting its container. Noble, and for decades, effective. But when the container is infinitely reproducible from a description, the lock on the container is decorative.

People built extraordinary things under the GPL. Entire operating systems, databases, web servers — the infrastructure the internet runs on, built for free, for the commons, because the license made sharing the default. Attention must be paid to what they built, even as the framework that protected it becomes obsolete. The work mattered. The license just can't do what it used to.

Two legal frameworks, built on opposite principles, destroyed by the same underlying shift: the thing they were attached to stopped being scarce.

What Remains?

If both the "you can't copy" framework and the "you must share" framework are dead, what's left?

The giants don't care. They already have the training data, the compute, the models. They're inside the gate. For them, copyright's death changes nothing — they bought, borrowed, or stole their way through while the buying was possible, and now they own the means of generation. Three thousand dollars a book was cheap.

For everyone else — every creator, every maintainer, every person who makes things and hoped the law would protect them — two centuries of legal infrastructure just evaporated. The "you can't copy this" framework is dead. The "you must share this" framework is dead. And nothing has replaced them.

That's where we are. Two frameworks down. No successor. And the enclosure didn't die with them — it moved somewhere the old tools can't follow.

This is part of a series exploring what happens when knowledge stops being property. Previous: Knowledge Was Always Free. Next: The Third Enclosure.

James Henry is a senior engineer who writes about AI, infrastructure, and what happens when the legal frameworks around knowledge stop working. He works with LLMs liberally — including in the writing of this post — because he believes the collaboration is the point. More at jamesrahenry.substack.com.