Anthropic's provenance system for Claude — an invisible, machine-readable watermark in generated text plus signed C2PA metadata on files, so content Claude had a hand in can be identified and verified.
Last verified · 2026-08-11 · by Moe Ameen
Claude Content Watermarking is Anthropic's system for marking content that Claude generates, detailed in its support documentation and surfaced in reporting on August 11, 2026. It has two parts. For text, new Claude models embed an imperceptible, machine-readable signal directly into the words — you cannot see it while reading, it does not change the meaning, quality, or readability of a response, and it stays attached when the text is copied and pasted or lightly edited. For files, Claude attaches signed provenance metadata using the Coalition for Content Provenance and Authenticity (C2PA) open standard, which records that Claude generated or handled the file and can reveal whether that provenance data was later altered.
The marking is applied at the model level to Claude models released on or after August 2, 2026, so it travels with the model rather than with any single interface. In practice that means it spans the Claude Platform API, the Claude apps, Claude Code, Claude Cowork, and Claude Tag, and supported models accessed through AWS, Google Cloud, and Microsoft Foundry. Anthropic says the markings apply worldwide, not only in the EU, and that it will publish technical details and tooling so users and third parties can check content for supported Claude marks.
The most important thing to understand before relying on it: the mark shows only that Claude had a hand in a piece of content, not that a Claude model wrote all of it. Because Claude can proofread, edit, or translate text a person authored, human-written work can carry the signal too — the mark is best read as "Claude processed this," not "AI wrote this." It is also fragile at the edges. The text signal weakens or disappears with heavy rewriting, paraphrasing, translation, or mixing with substantial human-written material, and a short passage may carry too little signal to detect reliably. C2PA file metadata can be stripped by format conversions, screenshots, or re-saving. Because of that, an unmarked piece is not proof a human made it.
The driver is regulation. The EU AI Act's Article 50 transparency requirements took effect on August 2, 2026 and push providers of generative AI to make synthetic output machine-readable and detectable; Anthropic signed the European Union's Code of Practice on transparency of AI-generated content and chose to apply the marks globally. Confirm the exact detection tooling and coverage on Anthropic's own documentation before building a compliance process on top of it.
Watermarking marks the draft; it does nothing to turn that draft into finished, published content — and that gap is where [Kompozy](/) works. Kompozy is a full AI content generation and multi-platform publishing engine, not a chat window, so a provenance-aware workflow looks concrete inside it. Your copy is generated under the [Persona Brief](/glossary/persona-brief) — voice, audience, and banned phrases enforced — so the text reads like you rather than median-prompt AI, which is what actually protects reach as detectors proliferate. Then Kompozy fans one source into [18 output formats](/glossary/output-buckets): [Blog Articles](/glossary/output-buckets), email newsletters, and text posts on the copy side, plus [Carousel Posts](/glossary/output-buckets) and quote cards rendered pixel-exact through [HyperFrames](/glossary/hyperframes). Each piece passes a per-post review gate before [Autopilot](/glossary/autopilot) schedules and publishes it across the eight social platforms plus blog and email, and that gate is exactly where you bake a consistent AI-disclosure line into whatever you ship.
Two honest, useful specifics. First, Kompozy runs on Claude alongside OpenAI for copy, so text it generates through Claude can carry the same provenance signal — the point is not to evade the mark but to build a disclosed, on-brand body of work where AI is a tool, not the tell. Second, Kompozy's images come from OpenAI's gpt-image and Google Gemini (for face-locked persona images), composited through HyperFrames — not from Claude — so Claude's C2PA "processed by Claude" tag does not attach to those assets; if you need C2PA provenance on an image, add it at your own render step rather than assuming it rode along. Knowing where the mark is and is not across your stack is the difference between a clean provenance story and a guessed one.
Two things. Text from new Claude models carries an invisible, machine-readable watermark embedded in the words, and files Claude generates or processes — including images like .png, .jpg, and .svg — get signed C2PA provenance metadata. Anthropic says it will publish tooling so users and third parties can detect supported Claude marks.
No. Anthropic is explicit that the mark shows only that Claude had a hand in the content, not that a Claude model wrote all of it. Because Claude can proofread, edit, or translate human writing, text a person authored can carry the signal too. Read it as "Claude processed this," not "AI wrote this."
The text signal weakens or disappears with heavy rewriting, paraphrasing, translation, or blending with substantial human-written material, and short passages may carry too little signal to detect. C2PA file metadata can be lost to format conversions, screenshots, or re-saving. That fragility cuts both ways: an unmarked piece is not proof a human made it.
It depends on which engine produced it. Kompozy uses Claude and OpenAI for copy, so text generated through Claude can carry the provenance signal. Its images come from OpenAI gpt-image and Google Gemini composited through HyperFrames — not Claude — so Claude's C2PA tag does not attach to those. The useful move is a clean disclosure habit rather than trying to strip marks.
The EU AI Act's Article 50 transparency requirements took effect on August 2, 2026 and push generative-AI providers to make synthetic output machine-readable and detectable. Anthropic signed the EU Code of Practice on transparency of AI-generated content and chose to apply the marks worldwide rather than only in Europe.