Every Claude model shipped on or after August 2, 2026 now marks its own output. Not metadata, not a header — a statistical signal baked into the words themselves, so it survives copy-paste out of the chat window and through some editing.
What it actually covers
This isn’t an app feature, it’s model-level. Claude, the API, Claude Code, Claude Cowork, Claude Tag, plus AWS, Google Cloud and Microsoft Foundry all inherit it, no toggle to flip. Generated .png/.jpg/.svg files get signed C2PA metadata instead. Models older than August 2 aren’t marked yet.
The trigger is the EU AI Act’s Article 50(2) transparency code, which Anthropic signed. But the marking ships globally, because running two output pipelines by region is dumber than running one.
The honest part
Anthropic says the quiet thing out loud: a detected mark means content passed through Claude, and that’s it. No mark proves nothing — edit heavily or paraphrase and it’s gone. So this is provenance plumbing, not an AI detector, and anyone selling it as one is wrong.
For enterprises there’s the Compliance API, now extended across Cowork and Claude Code in beta, letting security teams pull conversation and code-session records for audits and eDiscovery.
You Might Also Like
- Anthropic Just Launched Code Review in Claude Code and 54 of prs now get Real Feedback
- Claude Replay Turns Your Anthropic Claude Code Sessions Into Shareable Video Like Replays
- 13b Into Openai yet Microsoft Copilot Cowork Runs on Anthropic Claude
- Deepseek tui Tops Github Trending a Claude Code Clone Wired to Deepseeks api
- Alibaba Bans Claude Code for all Employees the Backdoor Fight With Anthropic Escalates
Top AI Product 
Leave a comment