Why Anthropic’s AI watermark is going further than its rivals — for now (www.businessinsider.com)

Anthropic has unveiled a new watermark for AI-generated text via its model, Claude, which aims for compliance with the EU's AI Act. This watermark is notable for its ability to remain detectable even when text is copied, translated, or extensively edited, raising questions about its necessity, as the EU regulations under Article 50(2) do not mandate such marking for standard edits. Some experts, like Caroline De Cock of the Lisbon Council, argue that Anthropic's method feels excessive, likening it to using a "sledgehammer" for a situation that could be addressed more delicately. This development is significant for the AI/ML community as it sets a high bar for transparency amid upcoming European regulations, prompting other tech companies like Google and OpenAI to develop their watermarking protocols. However, while Anthropic's initiative aims to enhance safety and minimize regulatory risks, it has sparked backlash from users who fear being misrepresented or having their authorship undermined. Critics point out that the watermark, which subtly influences Claude's word choices, might imply that the AI wrote the text, regardless of the user's original contributions. As AI watermarking becomes more prevalent, the debate around transparency versus authenticity in AI-generated content will likely intensify.