Anthropic is implementing machine-readable watermarks on Claude-generated text and files to comply with new EU regulations. These watermarks are designed to be imperceptible in text and digital signatures in media files. This means even content that has been merely cleaned up or translated by Claude could receive an AI signature. However, Anthropic acknowledges limitations, as proofread or lightly edited human content might falsely trigger a detection. Additionally, heavily rewritten or short texts may render watermarks undetectable. These changes aim to fulfill Article 50 of the EU AI Act, requiring disclosures for AI-generated or manipulated content. OpenAI is also working on compliance, focusing more on image and audio watermarking currently. This regulatory pressure is shifting the industry towards greater AI content identification. Other platforms like LinkedIn, Substack, and Snap are also enhancing their AI detection measures. All AI providers must bring their existing models into compliance with EU rules by December 2nd.