Anthropic Watermarks Claude for EU AI Act Compliance

Anthropic is now watermarking its AI-generated text from Claude to comply with new EU AI Act rules. This invisible, machine-readable mark travels with the text wherever it's pasted. This move comes as Europe's AI Act requires labeling of synthetic content. Companies can face fines up to 3% of global annual turnover for non-compliance. Anthropic is among the first major labs to embed this mark at the model level, affecting Claude, its API, and other platforms. Images also receive cryptographically signed provenance data. What's interesting is that within four hours of the announcement, a developer created a tool to remove the watermark. Others quickly followed, finding ways to strip or bypass it. Anthropic acknowledges these limitations, stating that heavily edited or translated text may not retain the mark. A text-detection API is planned to allow external testing of the watermark's durability. The bottom line is that AI transparency is shifting from disclosure to traceability, impacting how businesses will handle AI-generated content.