tech π€ Claude's secret text markings explained
Anthropic is implementing invisible text watermarks in Claude to follow Europe's AI transparency rules. This feature utilizes a version of Google DeepMind's open-source SynthID-Text system for detection. These marks are part of meeting obligations under the EU's AI Act, which requires machine-readable labels on synthetic content. The watermarks operate by subtly influencing word choices based on a secret key, leaving a pattern invisible to readers. This technology will not increase user costs or affect the quality of Claude's outputs. π€