Anthropic's AI watermark faces backlash for extending beyond EU requirements

Anthropic's AI watermark, which can appear in translated and edited text, has drawn criticism from some users and developers. The company claims the feature is designed to comply with the EU's AI Act, which requires AI-generated content to be marked in certain cases. However, experts argue that the watermark goes further than necessary, potentially marking even lightly edited text. Caroline De Cock, a senior fellow at the Lisbon Council, criticized the approach, calling it a 'sledgehammer' that targets simple proofreading and translation. Anthropic has stated it is rolling out the watermark globally, as it cannot apply it on a per-region basis. Some users have canceled their subscriptions, while developers have created tools to remove the watermark. The EU's regulations allow for AI outputs to remain unmarked if they are used for standard editing or do not substantially alter the original text. Anthropic's approach, which uses a detectable pattern to indicate the involvement of its AI model, has sparked debate over transparency and regulatory compliance.