Claude Will Now Secretly Watermark All Text and Images

Anthropic is rolling out invisible watermarks in Claude-generated text and digitally signed metadata in compatible file formats. This follows the company's commitment to the European Union's Code of Practice on Transparency of AI-Generated Content. Claude-generated text will embed a watermark that remains when copied or pasted into other applications and survives light editing.

Claude-created images, such as PNG, JPG, and SVG files, will carry signed provenance metadata to indicate processing by Claude and detect any later alterations. The watermark system is globally applicable for supported Claude models, including Claude, Claude Code, Claude Cowork, Claude Tag, and the Claude API. These models launched in the EU on or after August 2, 2026, will have the watermark technology built-in at launch.

However, Anthropic is retrofitting the capability into older models, though existing outputs may not yet display the watermark. The company plans to release tools and documentation for watermark detection. Users should note that the watermark may disappear if Claude-generated text is heavily edited, paraphrased, translated, or combined with other sources.

Very short passages might also lack a reliable watermark signal. Metadata can be lost during file conversion, re-saving, screenshot taking, or using metadata-stripping software. Anthropic notes that watermark detection only indicates Claude may have processed the content, not that Claude originally authored it. This transparency initiative has received support from around 190 organizations, including Anthropic, Google, Meta, Microsoft, and OpenAI, as outlined by the EU.