In a savvy move to keep up with E.U. regulations, Anthropic is rolling out machine-readable watermarks on text and images generated by its Claude models. These sneaky little markers are invisible to the human eye but allow platforms and people to identify AI-crafted content.
Anthropic’s watermarking wizardry involves embedding marks in text that stick around even if you copy, paste, or tweak the content a bit. This clever system covers all the Claude gang—Claude Platform (API), Claude Code, Claude Cowork, and Claude Tag—and extends to models running on AWS, Google Cloud, and Microsoft Foundry. Anthropic is also cooking up ways for users and third parties to detect these marks, with a how-to guide on the horizon.
This initiative aligns with the E.U. AI Act’s transparency requirements, which kicked off on August 2. New Claude models are watermark-ready from the get-go, while older models are getting the upgrade soon. The Verge highlights a four-month grace period for compliance, giving everyone a bit of breathing room.
However, Anthropic is playing it cautious. A watermark means content might have been touched by Claude, but it’s not a slam dunk. Claude could have been a helpful editor or translator, and the content might have been altered post-processing. An absent mark doesn’t guarantee human authorship either—heavy edits, short snippets, or older models can make marks vanish.
So, while Anthropic’s watermarking is a step towards transparency, it’s not foolproof. But hey, it’s a clever stride in the right direction for AI accountability!
Want to hear more? Join Mal & Matt on the Property AI Report Podcast each week!
Access from your preferred podcast provider by clicking here
Made with TRUST_AI - see the Charter: https://www.modelprop.co.uk/trust-ai