Superpower Daily
Army Cyber Puts AI Agents on Network Duty but Keeps Risk Decisi...AWS Brings GPT-5.6 Cross-Region Inference, With a Data-Location...Anthropic Gives Enterprise Teams Mythos 5’s Bug Hunt, Not Its P...Palo Alto and NTT DATA Build a $1B Security Alliance Without Ex...Cursor Gives Cloud Agents a Way to Come Back to the JobOrmat Bets AI Data Centers Will Make Geothermal Worth Drilling...Army Cyber Puts AI Agents on Network Duty but Keeps Risk Decisi...AWS Brings GPT-5.6 Cross-Region Inference, With a Data-Location...Anthropic Gives Enterprise Teams Mythos 5’s Bug Hunt, Not Its P...Palo Alto and NTT DATA Build a $1B Security Alliance Without Ex...Cursor Gives Cloud Agents a Way to Come Back to the JobOrmat Bets AI Data Centers Will Make Geothermal Worth Drilling...
Policypublished

A Byline on AI Copy Can Bring Liability Without Copyright, Scholar Says

A proposed authorship test distinguishes human creative choices from a platform contract’s separate allocation of output rights, input licenses and copyright risk.

By 2 min read
A Byline on AI Copy Can Bring Liability Without Copyright, Scholar Says
Sign in to save and react

Listen to this story

The audio brief

About 1:36
0:001:36
Read transcript
A proposed invisible watermark for Claude-generated text is exposing a separate legal problem: a byline may create responsibility without creating copyright. EUobserver reported that Anthropic is planning the watermark to signal Claude’s involvement, mainly for compliance with the EU AI Act. But that marker would show where text came from, not who owns it. Copyright scholar Daniel Gervais draws a sharper line. His proposed test asks whether a person made enough creative choices for copyright to attach. A LinkedIn post produced verbatim by Claude, he argues, could have no copyright owner. The same may apply to an article built from interviews, research, and source papers if the human contribution is mainly prompting and assembling material. Publishing either piece under a person’s name could still make that person accountable, and potentially liable, without giving them a transferable copyright. Contracts create a separate layer. In a review of 100 generative-AI agreements, John Newman and Andres Sawicki found that about two-thirds of contracts addressing output ownership named the user as owner, while preserving provider rights to use outputs. Providers received licenses to customer materials in 84.9 percent of contracts, and 69.8 percent made users pay when copyright claims reached the provider. Ninety-six percent let providers change terms without consent; 42 percent lacked meaningful notice requirements. Those researchers propose policy changes, not current law. The key constraint is that contractual ownership cannot settle whether human creativity qualifies for copyright in the first place.

Story brief

3 key points

Anthropic's reported plan for an invisible Claude watermark is about provenance for EU AI Act compliance, not ownership. Daniel Gervais argues that publishing model-generated work under a human byline may create accountability without creating copyright. That distinction matters because provider contracts can separately assign output rights and liabilities: in a review of 100 contracts, 84.9% licensed customer...

  1. 01

    EUobserver reported the watermark is intended to signal Claude text origin under EU AI Act compliance, not determine authorship or copyright.

  2. 02

    Gervais says verbatim Claude posts and heavily source-based generated articles may have no copyright owner, despite named-publisher liability.

  3. 03

    Among contracts explicitly addressing output ownership, about two-thirds named the user owner while retaining provider rights to use outputs.

An invisible watermark reportedly planned for Claude-generated text has brought a harder question into focus: who, if anyone, owns AI-written work? Copyright scholar Daniel Gervais argues that a byline can make someone accountable for generated text without giving them copyright or a right to transfer it. Platform contracts can still allocate separate rights and risks around the service.

EUobserver reported that Anthropic announced an invisible watermark for Claude-generated text in connection with EU AI Act compliance. The reported measure concerns the origin of generated text. Gervais’s proposed framework addresses a different issue in mixed human-AI work: whether the person made enough creative choices for copyright to attach.

Under Gervais’s interpretation, a LinkedIn post generated verbatim by Claude would have no copyright owner. He also says an article generated from interview recordings, research and source papers would not automatically belong to the person who prompted the model. Publishing either under a person’s name can create responsibility and potential liability, he argues, but not copyright or a transferable right.

John Newman and Andres Sawicki reviewed 100 generative-AI terms-and-conditions contracts in a working paper. Among contracts that explicitly addressed output ownership, about two-thirds named the user as owner while retaining provider rights to use outputs. The study found that 84.9% gave providers a nonexclusive license to customer materials, 63% allowed sublicensing, 30.1% allowed use for any business purpose, and 13.7% stated no limit on provider use. It also found that 69.8% required users to pay when their use led to copyright claims against a provider; 96% allowed providers to modify terms without consent, 88% gave that power to providers but not users, and 42% did not require meaningful notice of changes if notice was required at all.

Newman and Sawicki recommend voiding one-sided provisions and making firms, rather than individual users, legally responsible when copyrighted material is used without permission. Those are policy proposals, not current rules. Gervais’s framework likewise remains a proposed way to assess the difficult cases where people and AI both contribute to a work.

Sources

  1. euobserver.com[Interview] Does copyright protect your AI-generated content in Europe? Let’s find out
  2. equitablegrowth.orgTerms and conditions for generative AI users outline a new legal framework that puts U.S. consumers at risk

Related coverage

Continue reading