Anthropic DMCA'd a Developer for Changing One Word

Content creator Theo received his first DMCA strike this week. The violation? Changing one word in a markdown file in Anthropic's public Claude code repository.

The notice arrived while he was filming a video about an unrelated Claude code leak. "I woke up to a message from my assistant telling me I was DMCA'd by Anthropic and I just didn't believe it initially," Theo said on his stream. He grabbed his laptop to verify. The email from GitHub was real.

Here's what makes this interesting: the takedown wasn't for his YouTube coverage of the leak. It wasn't even for his fork of the actual leaked source code that he and a colleague had built and tested. The DMCA targeted his fork of Anthropic's official repository—the public one containing skills documentation and markdown files, not proprietary code.

The specific violation was a pull request where Theo had changed a single word in one file. That's it. One line of code modification earned him a copyright strike.

How We Got Here

The Digital Millennium Copyright Act exists to solve a specific problem that emerged in the late 1990s. Before the internet, distributing copyrighted material required physical media and distribution networks. The internet collapsed those barriers. Suddenly anyone with a DVD ripper could upload Avengers: Endgame to YouTube and reach millions.

The DMCA created what's called a "safe harbor" for platforms. If someone uploads infringing content to GitHub or YouTube, the copyright holder can't sue the platform—they have to go after the uploader. The platform's only obligation is to take down the content when notified. No investigation required. No judgment call about whether the claim has merit.

As Theo explained: "The goal of the DMCA was simple. Try to keep platforms from dying when there are different copyright holders trying to get their content taken off of them."

That speed and simplicity is by design. It's also why the system gets abused.

The Abuse Potential

The DMCA has been weaponized for decades. If you hold copyright on something—even something seemingly inconsequential—you can use it to take down content without proving actual harm.

Consider Nintendo and Wii emulators. Reverse engineering is a protected right in the United States. Making an emulator is fully legal. But Nintendo embedded a short encryption string—maybe 15 characters—into the Wii hardware. That string is necessary to decrypt game files from the disc into memory. Without it, games won't run.

Emulators like Dolphin need to include that string to function. Nintendo owns the copyright on those 15 characters. One tiny piece of the puzzle gives them leverage to shut down the entire project through DMCA claims, even though the emulator itself is legal.

Theo's case follows similar logic. He didn't steal code. He didn't redistribute proprietary material. He modified a public repository in the way developers modify public repositories every day. But Anthropic filed anyway.

The Corporate Response Gap

What struck Theo more than the DMCA itself was how it fit into a larger pattern. He's had substantially more positive experiences with OpenAI than Anthropic, despite having more personal contacts at the latter company.

"When things go wrong at OpenAI, people who work there come out and talk about it and do [something]," he said. "When things go wrong at Anthropic, the lawyers come out first."

He noted the contrast repeatedly: "OpenAI is weirdly more human, which is not what I expected and definitely not the direction I thought things would go."

This isn't about vilifying Anthropic. Theo himself backed off his initial anger after investigating further, saying "after doing more looking into it, talking to others, and figuring things out, I realized this is all much more complex than you might think."

But the observation stands. In an industry that trades heavily on community goodwill and developer advocacy, how a company handles mistakes matters as much as the technology they ship.

What This Actually Means

GitHub publishes every DMCA notice it receives. You can read them all. This transparency is useful—it creates a record of who's claiming what and whether patterns emerge.

Theo's notice sits in that public archive now, alongside thousands of others. Some legitimate. Some questionable. Some clearly abusive.

The system works exactly as designed. A company believes their copyright was violated. They file a notice. The platform removes the content. Fast, efficient, no judgment required.

The question isn't whether the system works. It's whether it works well. Whether one-word changes to public repositories should trigger the same mechanism as someone uploading leaked source code. Whether the safe harbor that protects platforms adequately protects developers.

Theo got his first strike for changing a word. He won't be the last person to discover that copyright law on the internet prioritizes speed over nuance.

The DMCA turned 25 years old recently. It was written for a world where the main threat was pirated DVDs. We're now arguing about whether forking a public repository and submitting a pull request constitutes infringement.

Something worth thinking about the next time you click that fork button.

—Bob Reynolds