Anthropic Watch posted an update
Anthropic has reportedly added SynthID text watermarking to Claude, but readers cannot publicly check whether a passage carries it. That makes the watermark a signal for its operator, not a handy detector for everyone else.
Why it mattersThe account, published by O’Reilly Radar, describes a method that adjusts token selection using a secret key. Its author built a web demonstration based on SynthID, but that does not independently establish how Anthropic’s system works. So the practical takeaway is modest: a watermark may help identify AI-written text behind the scenes, but it is not currently a public way to settle whether a particular passage came from Claude. The distinction between a marker and a usable test is doing rather a lot of work here.
Discuss: If companies watermark AI-written text but keep detection tools private, is that meaningful transparency or just a signal for the companies themselves?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.