To comply with the new transparency rules of the European AI Act, Anthropic now marks all texts produced by Claude with an imperceptible digital signature.
Anthropic has announced the deployment of an invisible, machine-readable watermarking system on texts generated by its recent Claude models, with a global rollout and not just in Europe. The brand remains attached to the text even when copying and pasting, and the company relies on the open C2PA standard to mark product files as well.
This initiative is a direct response to Article 50 of the European Union's AI Act, which became binding at the beginning of August, which requires providers of generative systems to clearly identify AI-produced content. Fines for non-compliance can reach €15 million or 3% of the company's global turnover. Anthropic has until December to extend this watermark to its older models.
Read the full article on the original source
Go to the source →