Anthropic Posts ‘How Claude Marks AI-Generated Content’ Without Explaining How Claude Marks AI-Generated Content
Anthropic support page: Anthropic has signed the EU AI Act’s Article 50(2) Code of Practice on Transparency of AI-Generated Content, as a provider of both generative AI models and generative AI systems. This article describes how we’re planning to put those commitments into practice, how marking works, and what its limitations are. We’ll update this article and publish more detailed technical…
Anthropic has committed to implementing the EU AI Act’s Article 50(2) Code of Practice on Transparency of AI-Generated Content, outlining its plans to mark generated content. When a Claude model generates text, it embeds an imperceptible watermark within the text itself, which remains undetected and does not alter the meaning, quality, or readability of the response.
This watermark will travel with the text when copied or pasted, potentially persisting through editing. Anthropic aims to enable users and third parties to detect these embedded watermarks and provenance metadata, but their claims of imperceptibility and its application to various file formats raise concerns. The company's claim that the watermark will be present regardless of the Claude product or surface is problematic.
Attempting to embed invisible characters within visible characters to mark AI-generated text raises questions about the feasibility and potential unintended consequences. If the watermark is visible, it contradicts Anthropic's assertion that it is imperceptible. Additionally, the process of embedding semantic detection for watermarking may lead to false-positive problems, as it could inadvertently affect the quality and readability of the text.
Written by urgent.news from Daring Fireball's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
