# Claude Adds Machine-Readable Marks

> Anthropic says supported Claude models will embed watermarks in text and attach signed provenance to supported files as EU AI Act transparency duties take effect.

_Source: Anthropic Help Center, European Commission guidance, and independent reporting · 2026-08-12 · 6 min read · Verified against primary sources_

Canonical: https://iyu.app/e/claude-text-watermark-eu-transparency

## The 60-second version

Anthropic is adding embedded text watermarks and signed file provenance to supported Claude output as EU transparency duties take effect.

**Key points**

- Supported generated text carries an imperceptible model-level watermark that can travel with copy and paste.
- Supported SVG, PNG, and JPEG files can carry signed C2PA metadata that records Claude processing and supports tamper checks.
- New models launched on or after August 2, 2026 support marking at launch, while support for older models is still being added.
- Coverage is worldwide across supported Claude products, APIs, and cloud access, but some file types and platforms may not support every mark.
- A detected mark suggests Claude processing; it does not prove original authorship, and no detected mark does not prove human origin.

**Verdict.** The policy creates useful provenance signals, but its operational value cannot be judged fully until Anthropic publishes detector access, accuracy, and robustness details.

## Full explainer


### The change — Two signals for two content types

Anthropic says supported Claude models will place an **imperceptible watermark inside generated text** and attach **digitally signed C2PA provenance metadata** to supported files. The text signal is applied at the model level; file marking depends on the format and platform supporting provenance metadata.

- **2** — complementary marking techniques
- **2 Aug 2026** — Article 50 transparency duties began applying
- **Worldwide** — coverage for output from supported Claude models


### Mechanism — How the marks travel

- **Generated text:** An invisible signal is woven into the text, travels with copy and paste, and may survive some editing.
- **Supported files:** SVG, PNG, and JPEG examples can carry signed C2PA provenance metadata that also supports tamper checks.
- **Product surfaces:** Supported models are covered across Claude, Claude Platform, Claude Code, Claude Cowork, and Claude Tag.
- **Cloud access:** Text marks extend to supported models through AWS, Google Cloud, and Microsoft Foundry; file provenance may vary.


### Interpretation — A mark is a signal, not an authorship certificate

A detected Claude mark indicates that content **may have been processed by Claude**. It does not establish who created the underlying ideas or whether the material changed later. An absent mark is equally inconclusive: older models, heavy editing, paraphrasing, translation, very short passages, unsupported workflows, and stripped file metadata can remove or weaken the evidence.

> **⚑ Caveat:** Anthropic has not yet published public technical details on detector accuracy, false-positive and false-negative rates, minimum reliable text length, or robustness under specific edits. Detection tools and fuller documentation are still forthcoming, so a result should not be used as a sole basis for enforcement or attribution.


### Regulation — What the EU framework requires

> Article 50 requires machine-readable transparency signals; it does not turn a watermark into conclusive proof of authorship.


## Primary sources

- [Anthropic — How Claude marks AI-generated content](https://support.claude.com/en/articles/16266773-how-claude-marks-ai-generated-content)
- [European Commission — Guidelines on transparency obligations for providers and deployers of certain AI systems](https://digital-strategy.ec.europa.eu/en/policies/guidelines-ai-transparency-obligations)
- [TechCrunch — Anthropic says it will watermark text generated by its AI models](https://techcrunch.com/2026/08/11/anthropic-says-it-will-watermark-text-generated-by-its-ai-models/)

---
_Published by iyu (https://iyu.app) — the day's AI news, checked against primary sources and rewritten in plain language. Free to quote with attribution and a link to the canonical URL._