AI-generated writing is becoming harder to distinguish from human work, and Anthropic is responding by putting an invisible identifier inside Claude’s output.
The company says supported new Claude models will embed machine-readable watermarks in generated text, while supported files can carry signed provenance metadata.
The change arrives as the European Union’s AI Act transparency obligations take effect, but Anthropic says its marking system will apply worldwide wherever supported Claude models are offered.
What Claude Will Mark
Anthropic says Claude models launched in the EU on or after August 2, 2026 support machine-readable marking from launch. The company is also working to add the capability to models released before that date, meaning the change should not be interpreted as an immediate watermark across every older Claude model.
For text, Claude embeds an imperceptible watermark directly into the generated content. Users will not see a label or special symbol, and Anthropic says the marking does not alter the meaning, quality or readability of the response.
The marking operates at the model level. Anthropic says it therefore applies across supported Claude products, including its API, Claude, Claude Code, Claude Cowork and Claude Tag. Supported models accessed through AWS, Google Cloud and Microsoft Foundry can also carry embedded text watermarks, although not every platform necessarily supports every type of marking.
Copy Paste Won’t Simply Remove It
One of the more significant aspects is how the watermark is designed to behave after generation. Anthropic says the mark travels with text when it is copied and pasted elsewhere and may persist through some editing.
The company has not, however, published enough technical detail to establish how much modification is required before a watermark becomes undetectable.
The watermark should not be treated as a definitive authorship test. A mark can indicate that content was generated or processed by Claude, but it does not establish that every idea, sentence or claim originated with the AI model.
Anthropic also says it plans to support users and third parties in detecting Claude’s marks, with more technical documentation to come.
EU Rules Raise The Stakes
The timing is closely connected to regulation. Article 50 of the EU AI Act’s transparency framework applies from August 2, 2026. Among other requirements, providers of certain AI systems must add machine-readable marks that enable AI-generated or manipulated content to be detected.
The European Commission’s Code of Practice says marking systems should be effective, interoperable, robust and reliable as far as technically feasible. The Code itself is voluntary, but the underlying transparency obligations in Article 50 are legally binding.
There is an important distinction here. EU rules explain why provenance has become a compliance issue, but Anthropic is not limiting its marking system to Europe. The company says supported-model markings will apply worldwide wherever Claude is offered.
Watermarking Becomes Industry Standard
Anthropic is entering a field where other major AI companies are already developing provenance technologies. Google DeepMind’s SynthID, for example, embeds imperceptible watermarks into AI-generated images, audio, video and text. For text generated through the Gemini app and web experience, SynthID adjusts token probability scores to create the watermark.
The broader business shift is significant. As AI-generated material becomes part of routine workflows in publishing, education, software and marketing, companies need ways to distinguish AI involvement from purely human production without relying solely on probabilistic AI detectors.
But watermarking does not solve the larger question of whether content is accurate, original or trustworthy. A watermark can provide information about provenance. It cannot establish whether a Claude-generated claim is true, whether a human substantially rewrote it, or whether the final material was responsibly edited.
Provenance Has A Bigger Test
Anthropic’s move therefore represents a shift from simply generating AI content to attaching information about its origin. For businesses, publishers and platforms, that could eventually make machine-readable provenance part of standard content workflows.
The technology still has to prove itself outside controlled generation. Anthropic’s own wording is deliberately cautious: the watermark travels through copying and may survive some editing, rather than surviving every transformation. Detection tools and technical documentation are also still being developed.
That makes the immediate significance less about suddenly being able to identify every AI-written sentence and more about establishing infrastructure for content provenance. If these systems become sufficiently reliable and interoperable, the invisible mark inside an AI-generated paragraph could become as routine as metadata attached to a digital file.
For now, Anthropic’s announcement marks another step toward that model: AI content may no longer need to announce itself visibly, but its origin could increasingly be designed to remain machine-readable.
Also read: US 100% Tariff Threat Over Russian Oil: Navarro Says Trump, Modi Will ‘Work It Out’
