Anthropic adds invisible watermarks

What's happened

Anthropic has begun embedding an imperceptible, machine-detectable watermark into Claude-generated text and files to comply with the EU AI Act’s Transparency Code. The mark travels with copied text, can persist through light editing, and will be applied globally to models released after Aug. 2. Users and developers have criticised the change and tools to remove marks have appeared online.

What's behind the headline?

What Anthropic has done

Anthropic has embedded a machine-readable watermark into Claude's text outputs and added file-level signatures for generated media. The watermark modifies low-stakes word choices so automated systems with the key can assign a probability that Claude processed the content.

Immediate consequences

• Publishers, schools and platforms will gain a new signal to investigate suspected AI content. This will increase the administrative load on content moderators and legal teams who must now interpret probabilistic marks.
• People who use Claude for proofreading, translation or light edits will risk carrying a provenance signal even when the human authored the underlying idea. That will create false positives.
• Developers have already built and published tools that attempt to remove or disrupt the watermark; those tools will push the problem into an arms race between watermarking and watermark-removal techniques.

Who benefits and who loses

• Regulators and platforms benefit because the marks create an auditable trail that supports compliance with the EU AI Act and platform policies.
• Anthropic reduces regulatory risk by applying the watermark globally and by offering a detection API, but it also concentrates a gatekeeping power: only parties with the detection key or API access will be able to verify the mark reliably.
• Ordinary users lose clarity: proofreaders, editors and non-native speakers who rely on AI tools for polishing text will face stigma or extra scrutiny.

Technical and legal trajectory

• Watermarks work by nudging model token selection. Light editing will not reliably remove the signal; a full rewrite will. That means actors who want to hide AI provenance will perform paraphrase or automated transformations, forcing detectors to evolve toward robustness or risk obsolescence.
• The approach will force courts, employers and platforms to define standards of evidence. A detected mark will not, by itself, prove authorship; heavy editing or mixed-source content will complicate attribution.

Forecast

• Expect a surge in third-party detection services and removal tools over the next months. This will push platforms and regulators to set detection and evidentiary standards and to clarify penalties for misrepresenting AI use.
• Content provenance will become a compliance and UX problem: companies will balance the legal safety of broad watermarking against customer churn and developer backlash. Anthropic will likely tune its approach and expand tool access to reduce friction, but the core trade-off—detectability versus convenience—will remain.

What you should watch

• Whether major platforms and publishers accept machine-readable watermarks as admissible evidence of AI use.
• How reliably detection APIs operate when text has been heavily edited, translated or mixed with human writing.
• The evolution of open-source tools that disrupt watermark signals and the countermeasures vendors adopt.

How we got here

The EU AI Act’s Transparency Code has required new models launched in the bloc after Aug. 2 to disclose AI-generated or edited content. Anthropic has implemented text-pattern watermarks plus file signatures and plans a detection API; Google and other firms have already been developing similar watermarking systems.

Our analysis

Business Insider UK (Aug 11–23, 2026) has provided the most detailed reporting on user backlash and technical concerns, quoting experts who warn Anthropic "is taking a somewhat broader interpretation" of the EU law and noting Anthropic's claim that the watermark "doesn't change the meaning, quality, or readability of Claude's responses." Business Insider also reported Anthropic will roll out a free API for detection and that the marks have been active on models released since Aug. 2. TechCrunch (Aug 11, 15, 12) explained the mechanics Anthropic is adopting — a token-selection pattern similar to Google DeepMind's SynthID-Text — and relayed Anthropic's line that "light editing probably won't remove the watermark completely" while a complete rewrite will. ZDNet (Aug 18, 2026) dissected the debate over whether nudging word choice is truly "low stakes" for writers and cited critics like John Gruber who fear watermarks could change phrasing choices. The Independent and New York Post (Aug 11–13, 2026) captured user reactions and community outrage, quoting the claim that the watermark could act like a "digital tattoo" and highlighting limits such as removal via heavy editing or screenshots for images. Combined, the coverage shows a split: vendors and regulators present watermarking as a practical compliance tool, while writers, some technologists and power users present it as an overreach that will produce false positives and an arms race with removal tools.

Go deeper

• Will platforms and publishers accept machine-detectable watermarks as definitive proof of AI use?
• How will Anthropic and rivals respond as removal tools spread on GitHub and social platforms?

More on these topics

• Anthropic - Artificial intelligence company

  Anthropic PBC is a U.S.-based artificial intelligence startup public-benefit company, founded in 2021. It researches and develops AI to "study their safety properties at the technological frontier" and use this research to deploy safe, reliable models for

• European Union

  The European Union is a political and economic union of 27 member states that are located primarily in Europe. Its members have a combined area of 4,233,255.3 km² and an estimated total population of about 447 million.

• Google - Technology company

  Google LLC is an American multinational technology company that specializes in Internet-related services and products, which include online advertising technologies, a search engine, cloud computing, software, and hardware.

• OpenAI - Artificial intelligence company

  OpenAI is an artificial intelligence research laboratory consisting of the for-profit corporation OpenAI LP and its parent company, the non-profit OpenAI Inc.

• LinkedIn - Website

  LinkedIn is an American business and employment-oriented online service that operates via websites and mobile apps. Launched on May 5, 2003, it is mainly used for professional networking, including employers posting jobs and job seekers posting their CVs.

• Reddit - Website

  Reddit is an American social news aggregation, web content rating, and discussion website. Registered members submit content to the site such as links, text posts, and images, which are then voted up or down by other members.

• TechCrunch

  TechCrunch is an American online publisher focusing on the tech industry. The company specifically reports on the business related to tech, technology news, analysis of emerging trends in tech, and profiling of new tech businesses and products.

• Substack - Internet company

  Substack is an online platform that provides publishing, payment, analytics, and design infrastructure to support subscription newsletters.

• Karoline Leavitt - American writer and political aide (born 1997)

  Karoline Leavitt (born August 24, 1997) is an American political spokesperson who has served as the 36th White House press secretary since 2025. A member of the Republican Party, she was the party's nominee in the 2022 election for New Hampshire's 1st...

• Meta Platforms, Inc. - Social media company

  Facebook, Inc. is an American social media conglomerate corporation based in Menlo Park, California. It was founded by Mark Zuckerberg, along with his fellow roommates and students at Harvard College, who were Eduardo Saverin, Andrew McCollum, Dustin Mosk

• India - Country in South Asia

  India, officially the Republic of India, is a country in South Asia. It is the second-most populous country, the seventh-largest country by land area, and the most populous democracy in the world.