Hear the biggest stories from the world of science | 14 August 2026
Anthropic deployed invisible watermarks to flag AI outputs and counter "AI slop"; researchers remain sceptical. The initiative reflects industry concern with data contamination from AI—adoption in research verification will determine effectiveness and compliance value.
Anthropic deployed watermarks on Claude output to meet EU AI Act transparency mandates. The feature's technical limitations—removability and inability to prove sole Claude authorship—undermine its effectiveness as an AI detection and governance tool.
Move induced by EU AI Act raises red flags for reliability and how content will be treated
Anthropic watermarked Claude text to comply with the EU AI Act via imperceptible word-choice patterns, applied globally. The watermark identifies only Claude involvement—not AI generally—and weakens on code, raising governance questions for enterprises.
Anthropic addresses concerns about its new watermark, which will affect Claude users globally. It said the watermark is to comply with the EU AI Act.