Innocent vs. Willful Infringement

US copyright law distinguishes between innocent infringement and willful infringement. The distinction matters enormously for statutory damages.

Under 17 U.S.C. 504, copyright owners who have registered their works can elect statutory damages in lieu of actual damages. The ranges are:

• Innocent

  $200 minimum per work

  Infringer did not know and had no reason to believe their acts constituted infringement

• Standard

  $750 to $30,000 per work

  Ordinary infringement, court determines amount

• Willful

  Up to $150,000 per work

  Infringer knew or had reason to know their acts were infringing

For a publisher with 10,000 articles in an AI training corpus, the difference between innocent and willful infringement is the difference between $2 million (innocent minimum) and $1.5 billion (willful maximum). Even at more moderate assessments, the difference is material.

How Embedded Provenance Creates Formal Notice

For infringement to be "innocent," the infringer must demonstrate that they had no reason to know the content was protected. When content carries embedded rights terms, that defense becomes difficult to sustain.

A C2PA manifest embedded in a text article includes, among other things, a rights assertion: a machine-readable statement of what licensing tier applies (Bronze for indexing, Silver for RAG, Gold for training). When an AI company's scraper ingests that article, the manifest is present in the scraped text.

The argument "we did not know the content was owned" fails when the content contained explicit ownership metadata. The manifest is not a metadata field that requires technical knowledge to interpret - it is structured data in a published open standard (C2PA) that any engineer can read with open-source tools.

Courts examining AI copyright disputes will need to address the notice question. Content that carries embedded, machine-readable rights terms presents a materially different notice argument than content that merely carries a byline and copyright notice in HTML headers.

EU AI Act and Machine-Readable Rights Reservations

The EU AI Act Article 4(3) specifically addresses machine-readable rights reservations. It requires providers of general-purpose AI models to implement policies that respect rights reservations made by content owners using machine-readable formats defined in the DSA (Digital Services Act) and its implementing regulations.

C2PA rights assertions are a machine-readable format. An AI company that ingests content with C2PA rights assertions and does not honor the licensing tier restrictions is potentially in violation of EU AI Act Article 4(3), in addition to any copyright claims.

The EU AI Act compliance angle provides a separate enforcement pathway from copyright law, and one with different remedies. Both pathways benefit from embedded provenance that documents what rights terms were present in the content at the time of ingestion.

The Evidentiary Weight of Cryptographic Proof

Evidence in copyright litigation typically involves documentation that can be challenged on authenticity grounds. A spreadsheet of publication dates can be questioned. A screenshot can be questioned. A log file can be questioned. These are not inherently fraudulent, but they depend on trusting the party that produced them.

A valid cryptographic signature is different in kind. Challenging a COSE signature requires demonstrating that the private key was compromised, the certificate chain is fraudulent, or the signature algorithm was broken. These are narrow and technically specific challenges that require expert evidence, not general skepticism about the producing party's motives.

Digital signatures have been accepted as evidence in courts globally. Electronic signature laws (ESIGN in the US, eIDAS in the EU) establish legal frameworks for digital signature admissibility. C2PA manifests are signed with standard cryptographic tools that satisfy these frameworks.

In practice, this means that a publisher presenting a C2PA evidence package has documentation that opposing counsel cannot dismiss as mere assertion. The verification is public, the algorithm is published, and anyone can reproduce the verification independently.

The Formal Notice Package

When a publisher discovers their signed content in an AI company's outputs, Encypher generates a formal notice package that documents:

• The original signed content with verified manifest (dated proof of ownership)
• The content found in the AI output (the allegedly infringing reproduction)
• The verification result (confirming the signed content is the source)
• The rights assertion in the manifest (documenting what licensing terms applied)
• The sentence-level attribution (specific sentences verified as matching)

This package is independently verifiable. The AI company's legal team can verify every claim in the package using open-source tools, without trusting Encypher. The verification is based on the C2PA open standard, not on proprietary assertions.

Publishers who use Encypher's notice feature report that many AI companies respond to notices by opening licensing discussions rather than litigating. The independently verifiable evidence package changes the economics of the dispute: litigation against cryptographic proof is expensive, while licensing is straightforward.

Registered vs. Unregistered Works

Statutory damages and attorney's fees under US copyright law are available only for registered works. Content provenance does not replace copyright registration - it is a different tool that strengthens the evidentiary record.

Publishers who want access to the full range of copyright remedies, including statutory damages, should register their works with the US Copyright Office. Provenance then provides the documentation that supports the willful infringement argument once the registration is in place.

For unregistered works, provenance still has value: it supports actual damages claims, licensing negotiations, and the willful infringement argument under the enhanced actual damages doctrine. It also applies in EU jurisdictions where copyright protection does not require registration.