Could someone train a detector for Claude’s watermark?
Anthropic's Claude uses a watermark that subtly biases token choices with a secret key, rather than adding visible characters. The challenge in detecting this watermark involves training a classifier to distinguish Claude's responses from other models, ensuring it identifies the watermark itself and not just Claude's writing style. A more robust approach would involve analyzing patterns in Claude's token choices when given repeated prefixes, to see if machine learning can approximate Anthropic's secret-key detector without knowing the key.
- Published
- Sep 7, 2026, 15:08
- Source type
- Dev community
- Tier
- Community
- Source status
- Healthy
Times shown in UTC
More details
Anthropic’s Claude watermark appears to work by subtly biasing which tokens Claude chooses using a secret key, rather than adding visible or hidden characters. The idea is to collect a huge number of Claude responses and responses from other models to the same prompts, then train a classifier to tell them apart. The main problem is making sure the classifier is detecting the watermark itself rather than just learning Claude’s writing style. A stronger experiment would repeatedly give Claude the same or similar prefixes and see whether a model can learn patterns in which next tokens Claude tends to choose. The interesting question is whether enough black-box examples could let machine learning approximate Anthropic’s secret-key detector without ever knowing the key.