Claude's AI Watermark Fades on Short, Edited, and Code Outputs — EU Compliance at Risk

Claude's AI Watermark Fades on Short, Edited, and Code Outputs — EU Compliance at Risk

Anthropic confirmed August 14 that Claude uses SynthID-Text, a probabilistic watermark whose signal weakens on short texts, edited passages, and code. A full rewrite erases it. The EU AI Act, effective August 2, mandates reliable AI-content identification. The planned detection API returns yes-or-no answers with variable confidence — a gradient, not a stamp.

Published

Read at another depth