Anthropic expands Claude watermarking despite a sampling loophole
The Decoder reports that Anthropic plans to add text watermarking to Claude Mythos 5, Claude Opus 4.7, Claude Opus 4.6, and Claude Sonnet 4.6, while Claude Opus 5.5 already includes the feature. On Opus 4.6 and Sonnet 4.6, the watermark is applied only when temperature, top_p, and top_k remain at their defaults, allowing API users to avoid it by changing any one setting. Anthropic describes the watermark as an imperceptible word-choice pattern that adds no tokens, leaves pricing and API formats unchanged, and barely reduces output speed, but the supplied report contains no independent measurements of these claims.
The rollout gives Anthropic a model-level watermarking mechanism across hosted platforms ahead of the EU AI Act deadline, but its coverage is conditional on sampling settings for two older Claude models.
Anthropic's watermark rollout leaves a loophole on two older Claude models
Matthias Bastian View the LinkedIn Profile of Matthias Bastian
Oct 31, 2026
Nano Banana Pro prompted by THE DECODER
Anthropic is expanding its text watermark to four more Claude models, but API users can dodge the mark on two of them by tweaking a single sampling parameter.
#Anthropic#Claude Mythos 5#Claude Opus 4.7#Claude Opus 4.6#Claude Sonnet 4.6#Claude Opus 5.5#Amazon Bedrock#Amazon Web Services#Google Cloud#Microsoft Foundry#Google DeepMind#SynthID