Anthropic extends Claude's text watermark to older models Nov 3
No opt-out, no public checker, and it marks the prose you had Claude edit too — all before the EU AI Act's Dec 2 deadline. Plus: server-side message threads.
Copy markdown
Four older models join the watermark this week
From Nov 3, Claude Mythos 5, Opus 4.7, Opus 4.6, and Sonnet 4.6 start embedding Anthropic's invisible statistical watermark in their text — across the API, Claude apps, Claude Code, and Cowork, worldwide (Bedrock may lag a few days). There's no toggle to turn it off.
It marks text you wrote, not just Claude's
The watermark is a pattern in word choices, so prose you had Claude draft or clean up carries it too. There's no public detector — checking runs through a private-preview API limited to regulators, researchers, and compliance teams, so you can't verify your own output.
Messages API gets server-side threads
A new beta (header message-threads-2026-08-12) keeps the conversation on Anthropic's servers, so each call sends only the new turn instead of replaying the full history. Lower token overhead and simpler state for anyone building chat or agent loops.
Fast mode for Opus 4.8 is on the clock
Anthropic deprecated fast mode for Claude Opus 4.8 on Oct 30; full removal lands Nov 30. If a latency-sensitive job pins that flag, migrate now — Opus 5.5 still offers fast mode, including on Google Cloud.