Anthropic Will Add Invisible Watermarks to Claude-Generated AI Text

Anthropic is introducing a new system designed to make AI-generated content easier to identify, with future and supported Claude models embedding invisible watermarks directly into generated text.

The company says the watermark will be machine-readable but imperceptible to users, meaning Claude's responses should look and read normally while still carrying a hidden signal indicating that the content was processed by Claude.

The move comes as AI companies face growing pressure to make artificially generated content more transparent.

Claude-generated text will carry hidden watermarks

Unlike a visible label saying "Generated by AI," Anthropic's approach places the watermark directly inside the generated text.

According to Anthropic, the watermark does not change the meaning, quality, or readability of Claude's response.

Because the watermark is embedded into the text itself, it can remain attached when someone copies and pastes Claude-generated content somewhere else.

Anthropic says the watermark may also survive some editing.

This could make it easier for platforms, publishers, schools, businesses, and other organizations to identify content that may have been processed using Claude.

Watermarking will happen at the model level

One important part of Anthropic's approach is that watermarking won't only exist inside the main Claude chatbot.

The company says the marking will happen at the model level.

That means supported Claude-generated text can carry the watermark across products including:

• Claude
• Claude Platform API
• Claude Code
• Claude Cowork
• Claude Tag
• AWS integrations
• Google Cloud
• Microsoft Foundry

For developers, this means applications using supported Claude models through the API may also receive watermarked text.

New Claude models will support marking from launch

Anthropic says Claude models launched in the European Union on or after August 2, 2026 will support machine-readable marking from launch.

The company is also working to introduce watermarking support for Claude models released before that date.

Although the changes are connected to European AI transparency requirements, Anthropic says marking will apply to supported Claude models wherever Claude is offered worldwide.

Why is Anthropic adding AI watermarks?

The rapid growth of generative AI has made it increasingly difficult to determine whether online content was written by a person, generated by AI, or created through a combination of both.

AI-generated articles, school assignments, marketing copy, social posts, emails, documentation, and even books can now closely resemble human writing.

Watermarking could provide another signal that helps determine where content came from.

Anthropic says greater transparency around AI-generated content can give people more useful context about the information they consume.

The company has also signed the EU AI Act's Code of Practice on Transparency of AI-Generated Content.

Anthropic is also working on watermark detection

Embedding a watermark is only useful if people can detect it.

Anthropic says it plans to provide users and third parties with ways to detect Claude's embedded watermarks.

More technical information about how detection will work is expected in future documentation.

A detected watermark could indicate that text was processed by Claude, but it won't necessarily prove that Claude originally wrote the entire piece.

For example, someone could write an article themselves and then use Claude to proofread, translate, summarize, or rewrite it.

The resulting text could still contain a Claude watermark.

The watermark won't be impossible to remove

Anthropic is also clear that AI watermarking has limitations.

Heavy editing could weaken or remove the detectable signal.

The watermark may become difficult to detect if Claude-generated content is:

• Heavily rewritten
• Paraphrased
• Translated
• Combined with other writing
• Extremely short

This means the absence of a watermark won't automatically prove that something was written entirely by a human.

Instead, watermarking should be viewed as another signal for identifying AI-generated or AI-processed content.

Claude-generated files will also carry provenance data

Anthropic's transparency plans extend beyond text.

For supported generated files such as PNG, JPG, and SVG files, Claude will use signed provenance metadata based on the C2PA standard.

This metadata can provide information about whether a file was processed by Claude and whether it may have been modified afterward.

The combination of text watermarking and file provenance could make AI-generated content increasingly traceable across the web.

AI-generated text is becoming easier to identify

Anthropic's move represents a significant change in how AI-generated writing may be handled.

Until now, users could copy text from an AI chatbot and paste it elsewhere without an official machine-readable signal traveling with it.

With model-level watermarking, that could begin to change.

The technology isn't perfect, and heavily modified text may still lose its detectable watermark.

But the direction is clear: as generative AI becomes increasingly common, AI companies and regulators are pushing toward a web where machine-generated content carries more information about where it came from.

For Claude users and developers, invisible AI watermarking could become an important part of how generated content is created, shared, and verified.