Skip to content
LiveNext 00:26:40
CoolingPolicy

Anthropic extends Claude text watermarking to older models, with a sampling loophole

Anthropic's help page says Claude text watermarks apply at the model level and lists models back to Opus 4.5, and that it signed the EU AI Act Article 50(2) transparency code. Detection is in private preview. The Decoder reports that on Opus 4.6 and Sonnet 4.6 the mark applies only at default temperature, top_p and top_k; Anthropic's page does not address this.

HEAT
OUTLETS
1
FIRST SEEN
LAST SIGNAL