SAN FRANCISCO, United States – Artificial intelligence safety research company Anthropic announced on Monday that new Claude AI models will embed an imperceptible digital watermark directly into generated text to track AI-produced content. The invisible marking technology integrates at the model level without altering the meaning or readability of the text, remaining attached when content is copied and pasted. The tool provides academic institutions and publishing houses with enhanced mechanisms to detect synthetic text following recent industry controversies, including author Jerry Falade and novelist Mia Ballard facing scrutiny over alleged AI use. However, Anthropic noted limitations, explaining that heavy editing, paraphrasing, or mixing outputs can degrade detection capability, while a detected watermark indicates AI processing rather than sole authorship. Anthropic becomes the second major AI developer to deploy text watermarking globally, joining Google DeepMind, which previously introduced its SynthID watermarking standard for Gemini-generated outputs.