Anthropic embeds invisible watermarks for Claude output globally: the era of AI content traceability officially begins Anthropic signs the EU AI Act Code of Practice, announcing that all newly released Claude models from August 2026 will embed content identification globally

Anthropic embeds invisible watermarks for Claude output globally: the era of AI content traceability officially begins Anthropic signed the EU AI Act Code of Practice, announcing that all newly released Claude models from August 2026 will embed content identification globally. This requirement is not limited to the EU market, but covers all Claude product lines including API, Claude, Claude Code, Claude Cowork and Claude Tag. Existing models will receive a legal transition period, but Anthropic said it is already working on retroactive adaptation of the watermark feature. Anthropic uses two identification mechanisms: the text output will carry an invisible watermark that does not affect the meaning and readability of the content. The watermark will still exist after copying and pasting, and "may be retained after a certain degree of editing." File formats such as images will have digital signature traceability metadata attached based on the C2PA open standard. The signature can indicate that the file has been processed by Claude and can reveal subsequent tampering. Text watermarks will also be available through cloud partners such as AWS, Google Cloud, and Microsoft Foundry, although these platforms may not support signature metadata. Anthropic plans to release verification tools in the future for users and third parties to detect these identifiers. Watermark detection has inherent limitations Anthropic admits that there are obvious limitations in watermark technology: detecting a watermark does not mean that the content is completely generated by Claude, because users often use Claude for proofreading, translation or summarization, and the output content may carry watermarks but the core ideas come from humans. Likewise, failure to detect a watermark does not rule out the involvement of AI—the model may have been output before the watermark feature came online, the text may have been heavily edited or translated, paragraphs may be too short to reliably detect, or metadata was stripped away in format conversions and screenshots. Whether a watermark can survive editing, reformatting and translation will be a key test of its true value. AI text detection has a particularly profound impact in socially sensitive fields such as education. Research shows that overreliance on AI tools can weaken critical thinking and writing skills, and scammers have used AI to register fake student identities to defraud financial aid. But unreliable detectors are equally dangerous and can lead to false accusations of cheating. If Anthropic's watermarking solution is more reliable than existing third-party detection tools, it may provide a new way out of this dilemma. Industry landscape: Different attitudes differ Anthropic is not the only one exploring watermarking technology