Anthropic's AI transparency efforts under EU rules are sparking debate
Experts are defending watermarking of AI-generated text against concerns the process degrades outputs after US giant Anthropic faced criticism for a transparency step that’s preferred under EU AI rules.
The bloc’s AI Act obliges in-scope developers to ensure models invisibly mark different forms of content – audio, image, video and text – to help web users spot synthetic content and shrink risks like disinformation.
With widespread access to generative AI tools, there’s also been a surge in low-quality content popping up online and elsewhere. Professional social network LinkedIn even launched a button last month that allows weary users to report suspected “AI slop”.
Anthropic – maker of the Claude chatbot – set out its approach to implementing the EU’s AI transparency code last week. Though its support page only mentioned “weav[ing] an imperceptible watermark directly into the text itself”, without further detail.
