Anthropic is expanding its text watermark to four additional Claude models — Mythos 5, Opus 4.7, Opus 4.6, and Sonnet 4.6 — with newer models like Opus 5.5 already carrying the mark and rollout on Amazon Bedrock potentially delayed by several days. The watermark, described as an imperceptible pattern in word choice that preserves meaning, quality, and readability, adds no extra characters or tokens, does not affect pricing or API formats, and negligibly impacts output speed; it also contains no user, organization, or conversation data. However, on the older Opus 4.6 and Sonnet 4.6 models, the watermark applies only when the three sampling parameters — temperature, top_p, and top_k — remain at their default values; altering any single parameter bypasses the watermark, a caveat Anthropic explicitly warns about for compliance purposes. This loophole does not appear to affect Mythos 5 or Opus 4.7. The retrofit is driven by the EU AI Act's December 2026 deadline, as models released after August 2, 2026, shipped with watermarking enabled, while older models are being updated in stages. The watermark operates at the model level across platforms including AWS, Google Cloud, and Microsoft Foundry and cannot be disabled by customers. Anthropic employs a variant of Google DeepMind's SynthID text method, which modifies word-selection randomness to create a statistically detectable pattern; a detection API is in private preview. OpenAI's textGrain system, launched October 2026, activates by default only for EU ChatGPT and Codex users, while API customers worldwide can toggle it on or off. Critics argue watermarking may degrade semantic precision by substituting synonyms, and layered watermarks in legal documents could create compliance complications.