Anthropic has announced that it will soon apply invisible watermarks to all content processed or generated by its AI models. This move is in compliance with the European Union AI Act, which mandates AI system providers to label text, audio, image, and video outputs.
According to the company, these watermarks are machine-readable but will remain invisible to human users. Unlike the EU law's requirement, which only emphasizes marking significant changes to content, Anthropic has adopted a more stringent approach and intends to apply these watermarks even for minor edits and simple grammatical corrections.
These changes are scheduled to be implemented from day one for all new models released globally (not just in Europe). Furthermore, other files produced by this AI will contain digitally signed metadata to authenticate the origin and source of the work.