DMCA

The DMCA — spelled out, Digital Millennium Copyright Act — is a law of the United States that took effect in 1998. It establishes what rights creators have on the internet and what obligations platforms carry when users can upload their own content. The law came about because copyright law at the time did not yet account for the internet. It was meant to give a clear answer to questions like: Who is responsible when someone posts a music video without permission on a website — the user or the platform? The DMCA formally applies only in the US, but it has worldwide effect because the largest platforms — YouTube, GitHub, Google — are headquartered there and the law is therefore applied to them.

Liability protection as the foundation of the modern internet

The most important part of the law is the so-called safe harbor. It states: a platform is not automatically liable for copyright-infringing content posted by its users — as long as it acts quickly once notified. Without this provision, every platform would be responsible for everything its users upload. That would have made the internet as we know it practically impossible.

The safe harbor is so influential because it describes a deal: platforms don’t have to review every piece of content in advance, but they must respond when violations are reported. That sounds like a compromise — and it is. Critics on one side say this turns platforms into tools for copyright infringement. Critics on the other side say rights holders abuse the system to suppress legitimate content.

The takedown procedure step by step

At the core of the DMCA is a formal notification procedure. A rights holder — for example, a music label — discovers a video on YouTube that contains their song. They send a written notice to YouTube, called a takedown notice. In it, they must state which work was infringed, exactly where the content can be found, and that they are indeed the rights holder.

YouTube is then obligated to block the content. The user whose video was taken down can file an objection — this is called a counter notice. In it, they explain why they believe they used the material legally, for example because they hold a license or because the use falls under the American legal concept of fair use. Fair use permits the use of content without permission in certain cases, such as for criticism, parody, or education. After that, the rights holder has two weeks to file a lawsuit — if they fail to do so, the content is restored.

The system works cleanly in theory. In practice, takedown notices are sent out en masse and sometimes automatically, even when no actual infringement exists. According to Google’s Transparency Report, more than 900 million removal requests for search results were received in 2023 alone — a large portion of which were erroneous or abusive.

DMCA in tech news and AI debates

The term regularly appears in the news whenever well-known content gets taken down: a Twitch stream playing music in the background; a GitHub repository that gets suspended because it allegedly contains copyrighted code; a YouTube video that briefly shows a TV broadcast during a livestream. For developers, streamers, and content creators, the DMCA is part of everyday work.

The law is especially relevant in the AI debate right now. Several lawsuits target AI companies that trained their models on copyrighted text, images, or code. The central question: Does this count as a copyright infringement — and does the DMCA even apply here? The law was written in 1998, long before machine learning existed. Whether it applies to AI training is currently being decided by courts in the US, and the outcome will affect the entire AI industry.

The law also became relevant for platforms like GitHub when the AI assistant Copilot was accused of reproducing copyrighted source code during code generation. Such cases show: the DMCA is not a dusty law from the early days of the internet, but an active battleground where the future shape of copyright in the age of AI is being decided.