Anthropic says text watermarking scheme relies on inconsequential words

And other AI model makers are expected to deploy something similar