Anthropic’s New Watermarks for Claude-Generated Content: What You Need to Know

Anthropic’s New Watermarks for Claude-Generated Content: What You Need to Know

In response to the E.U. AI Act’s regulations regarding transparency in AI-generated content, particularly those found in Article 50 of the E.U. AI Act, Anthropic announced this week that it would begin embedding machine‑readable watermarks in the output of all newly released Claude models.  Anthropic’s announcement created significant waves across businesses and their employees by adding that these watermarks would be included in the text, and would survive copying and pasting of said text across mediums, meaning that Word documents or similar text files into which Claude-generated text is pasted will include these watermarks indicating that such text was generated using AI.  Further, Anthropic will apply this new policy globally, meaning that even users outside of Europe will find their Claude-generated content contains these embedded watermarks (or, in the case of non-text files, metadata indicating their provenance). Claude models launched on or after August 2, 2026, will support machine-readable watermarking at launch, while Anthropic is working to add marking support to models released before that date.  Although limitations remain, such as watermarks potentially being removed via heavy editing, paraphrasing, translation, or certain format conversions, this has significant implications for the usage of AI-generated text going forward.

‍

Why it matters under U.S. law

Although meant to comply with the E.U. AI Act, Anthropic’s watermarks could also be considered a preemptive signal to the U.S. Federal Trade Commission (FTC) of the company’s intention to meet truth‑in‑advertising and transparency expectations already enforced by the FTC under Section 5 of the FTC Act, which prohibits unfair or deceptive acts or practices. The FTC has emphasized that deceptive AI marketing, hidden AI‑generated reviews, and undisclosed automation could potentially be violative of Section 5 of the FTC Act, and Anthropic’s watermarks could support provenance programs and help substantiate claims regarding the creation and moderation of content. Additionally, where a watermark or provenance tag qualifies as “copyright management information” (CMI), its unauthorized removal or alteration may give rise to liability under the Digital Millennium Copyright Act (DMCA), which broadly protects CMI and, when removed with the actual or constructive knowledge that its removal or alteration will facilitate copyright infringement, prohibits the intentional removal or alteration of CMI.

‍

Practical Takeaways for Businesses

In light of Anthropic’s announcement, businesses using, or contracting with vendors who utilize, AI should begin taking the following steps to protect their interests:

• Protect the integrity of provenance markers through contractual safeguards. In order to ensure that the provenance of vendor-supplied material can be verified for either business or, where potential intellectual property infringement presents material risk, legal reasons, agreements with agencies, vendors, freelancers, and other content creators should expressly prohibit the unauthorized removal or alteration of watermarks, provenance metadata, and similar identifiers.

‍

• Establish detection, verification, and recordkeeping procedures regarding content provided both internally and externally. Additionally, businesses should implement processes to verify Anthropic watermarks and other provenance markers both when content is received and before it is published or distributed.  Logging these processes can help to preserve a reliable record for internal investigations and facilitate responses to inquiries from platforms, regulators, or other third parties.

‍

The E.U. AI Act presents a looming or, in the case of businesses operating in the E.U., present regulatory reality, and given the significant legal and operational requirements for various types of AI systems under the E.U. AI Act, proactive steps should be taken to ensure that, whenever your business becomes subject to the E.U. AI Act, you will be ready to meet your compliance requirements.  PAG Law’s Data Privacy, Cybersecurity, and AI Governance practice is ready to assist with your compliance journey. 

‍

^{Zachary Soto is a Partner at PAG Law PLLC, where he chairs the Privacy Law, Cybersecurity, and AI Governance practice. He has practiced privacy, cybersecurity, and corporate transactional law for over 16 years and is certified by the International Association of Privacy Professionals as a Certified Information Privacy Professional (CIPP/US) and AI Governance Professional (AIGP).}

‍

^{Disclaimer: This publication is provided by PAG Law PLLC for general informational purposes only and does not constitute legal advice or create an attorney-client relationship between PAG Law and the reader. The content reflects the views of the author as of the date of publication and may not reflect subsequent developments in law, regulation, or policy. Readers should not act or refrain from acting on the basis of any information contained herein without seeking professional legal counsel tailored to their specific circumstances and jurisdiction. PAG Law expressly disclaims all liability with respect to actions taken or not taken based on any or all of the contents of this publication. This material may be considered attorney advertising in some jurisdictions.}