Anthropic's watermark rollout leaves a loophole on two older Claude models
| Source: THE DECODER
Tags: Anthropic, Claude Opus 4.6, Claude Sonnet 4.6, Claude Mythos 5, EU AI Act, Watermarking, Amazon Bedrock
Anthropic is extending its text watermark to four more Claude models ahead of the EU AI Act's December deadline, but on Opus 4.6 and Sonnet 4.6 changing any sampling parameter disables it.
Details
In an email to API customers, Anthropic said it will watermark Claude Mythos 5, Claude Opus 4.7, Claude Opus 4.6 and Claude Sonnet 4.6. Newer models such as Claude Opus 5.5 already carry the mark; the Amazon Bedrock rollout may lag by a few days. Anthropic describes the watermark as an imperceptible pattern in word choice that adds no tokens, does not change pricing or formats, barely affects speed, and carries no user or conversation data. On Opus 4.6 and Sonnet 4.6, the mark applies only when temperature, top_p and top_k are left at defaults, so a small tweak avoids it. Anthropic only warns customers relying on it for compliance to keep defaults. A detection API is in private preview.