Anthropic is going to watermark everything Claude writes. You will not be able to see it, and neither will existing AI detectors. The fear was that marking the text would wreck the prose. The reality is… complicated. @AnthropicAI is using a version of @Google's SynthID: A secret key picks between words that were already about equally likely, so the odds Claude gives each word hold on average. John Gruber called it “a perversion of writing.” Our own @hammer_mt went in scared, read how it…