The DMCA, or Digital Millennium Copyright Act, is the United States law that adapted copyright to networked computers. Signed on October 28, 1998, it does two main things. It shields online services from liability for what their users post or link to, provided they remove infringing material when a rights holder complains, a regime known as notice-and-takedown. And it makes it illegal to break, or sell tools that break, the digital locks placed on copyrighted works. The second part, written with DVD encryption in mind, has become the weapon of choice in lawsuits about scraping search results and training artificial intelligence (AI) models.

How notice-and-takedown works

The safe harbours sit in Section 512 of Title 17 of the US Code. They cover four activities: transmitting data (512(a)), caching it (512(b)), hosting material stored at a user's direction (512(c)) and linking to it through search engines or directories (512(d)).

Qualifying has conditions. The provider must register a designated agent with the US Copyright Office, adopt and reasonably implement a policy for terminating repeat infringers, accommodate "standard technical measures", and act expeditiously once it has actual knowledge of infringement or awareness of facts that make it apparent, the so-called red flag test. It also loses protection if it receives a financial benefit directly attributable to infringement it has the right and ability to control.

A valid notice under 512(c)(3) has six elements: a physical or electronic signature, identification of the copyrighted work, identification of the infringing material with enough information to locate it (typically a URL), contact details, a statement of good-faith belief that the use is unauthorised, and a statement, under penalty of perjury, that the sender is authorised to act for the owner. The provider removes or disables the material and tells the uploader.

The uploader can then send a counter-notice under 512(g). Unless the complainant files a lawsuit, the provider restores the material within 10 to 14 business days. Section 512(f) makes knowing misrepresentation in a notice actionable, and 512(h) lets a rights holder obtain a subpoena from a court clerk, without filing a lawsuit, to unmask an alleged infringer.

In practice the process is industrial. Google's Transparency Report passed 10 billion URLs requested for removal from Search in November 2024, from more than 600,000 copyright owners, according to figures reported at the time.

The anti-circumvention rules

The second pillar is Section 1201. Its subsection (a)(1)(A) bars circumventing a "technological measure that effectively controls access" to a copyrighted work. Subsection (a)(2) bars trafficking in tools or services that do so, and 1201(b) bars trafficking in tools that defeat copy controls. Section 1202 protects copyright management information (CMI), such as author names and copyright notices attached to a work, and forbids removing or falsifying it.

Penalties are set in Section 1203. Statutory damages run from $200 to $2,500 per act of circumvention and from $2,500 to $25,000 per CMI violation. Willful commercial violations are also crimes under Section 1204. Because damages multiply per act, exposure for a high-volume scraper can be vast.

Section 1201 has a safety valve. Every three years the Librarian of Congress, on the Copyright Office's recommendation, grants temporary exemptions, covering uses such as phone unlocking, security research and repair. The tenth triennial proceeding opened on June 9, 2026, with petitions due August 24, 2026; exemptions granted in it would run from October 2027.

Origin and evolution

The DMCA implemented two World Intellectual Property Organization (WIPO) treaties adopted on December 20, 1996: the Copyright Treaty and the Performances and Phonograms Treaty. Both required legal protection against circumventing technical locks. Section 512 began life separately, as the Online Copyright Infringement Liability Limitation Act, a compromise between telecoms companies and rights holders. It became Title II of the DMCA.

Courts shaped the rest. Field v. Google, decided on January 19, 2006, applied the caching harbour to Google's cached pages, a ruling covered in the PPC Land explainer on noarchive. Viacom v. YouTube, a $1 billion claim filed in 2007, produced the Second Circuit's April 5, 2012 ruling that only knowledge of specific infringing clips defeats the harbour; the parties settled in 2014. In Lenz v. Universal, the "dancing baby" case, the Ninth Circuit held in 2015 that senders must consider fair use before issuing a notice. And in BMG v. Cox, the Fourth Circuit ruled on February 1, 2018 that Cox lost its safe harbour by failing to terminate repeat infringers.

The Copyright Office's Section 512 study, published on May 21, 2020, concluded that the balance Congress intended had become "unbalanced", with rights holders facing an endless "whack-a-mole" of reuploads.

Why it matters for marketers and publishers

For search marketers, takedowns are a ranking variable. Removed URLs disappear from Google Search, and the process can be abused. In March 2026, Google removed two articles investigating Clickout Media after DMCA complaints alleging they copied a 2024 Verge article; Google reinstated them on March 31, faster than the counter-notice timetable would allow.

For publishers, the DMCA is one of the few statutes that reaches AI training directly. Section 1202 claims argue that stripping bylines and copyright notices from articles during dataset preparation is a separate violation, carrying statutory damages without proof of lost sales. Section 1201 claims turn bot defences into legal barriers. On December 19, 2025, Google sued SerpApi under Section 1201, alleging that it circumvented SearchGuard, the JavaScript challenge Google introduced in January 2025. Reddit had already sued SerpApi, Oxylabs, AWMProxy and Perplexity on October 22, 2025, using a test post visible only to Google's crawler to show its content had been lifted from Google results.

Limits and disputes

The statute is contested on both sides. Rights holders say the 512 harbours let platforms profit from piracy while shifting enforcement costs onto creators. Civil liberties groups say notices are cheap to send, rarely penalised under 512(f) and used to suppress criticism.

Section 1201 draws heavier fire. It lacks a general fair use defence, and courts disagree on whether a violation needs any link to infringement. The Federal Circuit required such a nexus in Chamberlain v. Skylink in 2004; the Ninth Circuit rejected it in MDY v. Blizzard in 2010.

The scraping cases have now produced a split. On July 20, 2026, Chief Judge Yvonne Gonzalez Rogers dismissed Google's claims against SerpApi, holding that "SearchGuard cannot be said to effectively control access to a work protected under the Copyright Act" for results containing no copyrighted content. Eleven days later, on July 31, Judge Paul Engelmayer in New York let Reddit's core anti-circumvention claims proceed, though he dismissed its 1201(b) count. SerpApi had argued that SearchGuard cannot be "effective" when any human can reach the same pages, and called Google "the largest scraper on the planet".

Section 1202 claims against AI developers have mostly failed. Judge Colleen McMahon dismissed Raw Story and AlterNet's case against OpenAI on November 7, 2024 for lack of standing, finding no concrete injury from removal of CMI without dissemination. On September 16, 2026, the Ninth Circuit affirmed dismissal of the CMI claims over GitHub Copilot, with Judge Eric Miller writing that "one who creates a new work and fails to include CMI cannot be said to have 'removed' or 'altered' anything."

Not the same as

Copyright infringement itself is a separate claim under Section 501, with statutory damages of up to $150,000 per work for willful infringement, as the PPC Land explainer on statutory damages sets out. Fair use rulings on AI training, such as the June 2025 decision that preceded the $1.5 billion Anthropic settlement, are copyright rulings, not DMCA ones.

The EU Copyright Directive (2019/790) is often called Europe's DMCA, but it works differently. Article 17 makes large content-sharing platforms directly liable unless they make "best efforts" to license works and prevent uploads, a stricter standard than takedown after notice. Its Article 4 lets rights holders opt out of commercial text and data mining in machine-readable form.

The Digital Services Act (DSA) notice-and-action mechanism, mandatory for all EU intermediaries since February 17, 2024, covers all illegal content, not just copyright, and adds trusted flaggers and statements of reasons. It has no counter-notice clock or anti-circumvention rules.

A legal removal from Google is broader still, covering defamation, privacy and court orders, and is distinct from spam-policy manual actions.

Recent developments

The fight over Section 1201 has moved to the amended Google complaint. Google refiled on August 10, 2026, citing its Reddit licence and two unnamed partners as authority for SearchGuard. SerpApi moved to dismiss again on August 24, noting that none of those contracts had been filed. The hearing set for today, October 13, 2026, was vacated on October 5, and the court said it would reset it or decide on the papers, according to the court docket. In New York, SerpApi's answer to Reddit is due on October 23, 2026.

Section 1202 survives in the publishers' case against OpenAI and Microsoft. The parties cross-moved for summary judgment on September 4, 2026, over more than 10.8 million asserted works, with OpenAI arguing that any CMI removal was an incidental by-product of text extraction.

Congress, meanwhile, is borrowing DMCA machinery. The TRAIN Act, introduced on January 22, 2026, would let copyright owners obtain clerk-issued subpoenas modelled on Section 512(h) to learn whether their works trained an AI model. And on March 25, 2026, the Supreme Court ruled for Cox in Cox v. Sony, holding that mere knowledge a service will be used to infringe does not establish contributory liability, which lowers the stakes for providers that fall outside the safe harbours.

Timeline

  • December 20, 1996: WIPO adopts the Copyright Treaty and the Performances and Phonograms Treaty
  • October 28, 1998: President Bill Clinton signs the DMCA into law
  • October 2000: First triennial Section 1201 exemptions take effect
  • January 19, 2006: Field v. Google applies the Section 512(b) caching harbour to search caches
  • March 2007: Viacom sues YouTube for $1 billion
  • April 5, 2012: Second Circuit rules in Viacom v. YouTube on specific knowledge
  • September 14, 2015: Ninth Circuit rules in Lenz v. Universal that notices must consider fair use
  • December 1, 2016: Copyright Office launches its electronic designated agent directory
  • February 1, 2018: Fourth Circuit denies Cox the safe harbour over repeat infringers in BMG v. Cox
  • April 17, 2019: EU adopts the Copyright Directive, including Article 17
  • May 21, 2020: Copyright Office publishes its Section 512 report
  • February 17, 2024: DSA notice-and-action rules apply to all EU intermediaries
  • November 7, 2024: Raw Story's Section 1202 case against OpenAI is dismissed for lack of standing
  • November 2024: Google's copyright removal requests pass 10 billion URLs
  • October 22, 2025: Reddit sues SerpApi, Oxylabs, AWMProxy and Perplexity under Section 1201
  • December 19, 2025: Google sues SerpApi under Section 1201
  • January 22, 2026: TRAIN Act introduced in the House
  • March 25, 2026: Supreme Court decides Cox v. Sony
  • March 30-31, 2026: Google removes, then reinstates, the Clickout Media articles
  • June 9, 2026: Tenth triennial Section 1201 rulemaking begins
  • July 20, 2026: Google's Section 1201 claims against SerpApi dismissed
  • July 31, 2026: Reddit's core Section 1201 claims survive dismissal
  • August 10, 2026: Google files its amended complaint
  • September 4, 2026: Summary judgment motions filed in the OpenAI news publisher litigation
  • September 16, 2026: Ninth Circuit affirms dismissal of Copilot CMI claims
  • October 5, 2026: Hearing on SerpApi's second motion to dismiss vacated

Summary

Who: Copyright owners and their enforcement agencies, online service providers such as Google, YouTube and internet access providers, the US Copyright Office, and, increasingly, platforms like Google and Reddit suing scrapers, and publishers suing AI developers.

What: A 1998 US law that grants online services safe harbours from copyright damages in exchange for notice-and-takedown, and that prohibits circumventing technical access controls and stripping copyright management information.

When: Signed on October 28, 1998, implementing 1996 WIPO treaties; reshaped by court rulings from 2006 to 2026, with exemptions renewed every three years and the tenth cycle under way since June 9, 2026.

Where: The United States, through Sections 512 and 1201 to 1205 of Title 17, though its takedown process governs global platforms; the EU runs parallel systems under the Copyright Directive and the DSA.

Why: Congress wanted the internet to grow without holding intermediaries liable for every user upload, while giving rights holders legal backing for digital locks. Today the same provisions decide which URLs vanish from search and whether scraping and AI training can be fought outside ordinary copyright law.