4 ms·
> Honest note: Anthropic has not shipped a public Claude watermark detector yet. This tool uses rewrite-based neutralization — a meaning-preserving paraphrase w
by arcfour 2mo ago
> Honest note: Anthropic has not shipped a public Claude watermark detector yet. This tool uses rewrite-based neutralization — a meaning-preserving paraphrase with a non-Claude model — which is the attack path watermark research points to. Not affiliated with Anthropic.
Well, they should have run their own AI slop website through their tool...
- nunez 2mo agoThis attack was actually pointed out in the watermarking paper linked above. The researchers added an instruction to the prompt that switches letters like a Caesar Cipher. It lowers the quality of the output from the LLM but alters the "red list" enough for a watermark detection tool to fail at detecting the watermark.
- fl0id 2mo agoalso their example for rewriting just completely changes it. might as well redo it in this case (with another model or by hand)