Anthropic Rolls Out AI Watermarks to Meet EU Regulations
The AI company is implementing imperceptible watermarks in its content to comply with the European Union's AI Act transparency requirements.
Anthropic, the artificial intelligence company behind the Claude chatbot, has begun embedding imperceptible watermarks into its AI-generated content to align with the European Union's AI Act. The law, which took effect on August 2, mandates transparency measures to help users distinguish between human-created and AI-generated material.
According to Anthropic's support website, these watermarks are automatically integrated into text, images, and files produced by supported Claude models. The company states that the watermarks are designed to be undetectable to the human eye and do not affect the quality or readability of the content. They are intended to persist even when the text is copied, pasted, or moderately edited.
The watermarking implementation will be applied globally across all Anthropic offerings, including the Claude Platform API, Claude Code, Claude Cowork, and Claude Tag.
Anthropic has noted that its detection system has limitations and may not always successfully identify AI-generated content, particularly if the text has undergone substantial editing, paraphrasing, translation, or if the passage is very brief. The European Commission has reported significant industry support for its transparency initiatives, with nearly 200 companies committing to comply with the AI Act's rules as of July 31.
Major technology firms, including OpenAI, Meta, Google, and Microsoft, have also signed onto similar pledges. The European Union has stated that these commitments signal an intention to foster public trust in AI and to reduce deception and misinformation.
The move by Anthropic and other AI developers comes amid growing concerns about the proliferation of low-quality, AI-generated content, often referred to as "AI slop." This issue gained attention recently when a debut novelist's book deal was reportedly canceled due to publisher concerns about the use of AI in manuscript creation.
However, Anthropic's new watermarking policy has met with some criticism from users who view it as ineffective or unnecessary. Some commenters on social media platforms have pointed out that readily available open-source models without watermarks could be used as alternatives by those seeking to bypass such measures, questioning the overall impact of Anthropic's approach.