#claude#anthropic#marca de agua#ia

    Claude now puts an invisible watermark in its text: what it proves and what it does not

    Claude puts an invisible watermark in the text it writes. What it does, what it cannot prove, and why it is a clue, not a verdict.

    The invisible watermark in Claude is a signal Anthropic documented on November 3, 2026 that travels inside the text Claude writes, even if you copy and paste it. It can be detected, but only in a private preview, and it isn't proof: if the text is translated, heavily edited, or very short, it can lose it.

    What is Claude's invisible watermark

    Anthropic documented Claude's text watermark on November 3, 2026, in the Claude Platform release notes and its Help Center. The underlying rule is simple: text without the mark doesn't prove a person wrote it. The mark can show up, but its absence doesn't guarantee anything.

    Claude Platform release notes, where Anthropic documented the text watermark

    What the watermark does

    Three things stand out:

    • It travels inside the text, even when you copy and paste it elsewhere.
    • It survives some edits.
    • It can be detected, but only in a private preview: today it isn't an open tool for everyone.

    What the watermark can't prove

    The mark isn't proof. If you translate the text, edit it heavily, or it's very short, it can lose it. So instead of accusing someone, treat it as a clue, not a verdict. And the absence of the mark doesn't show that a person wrote the text either.

    Two people working on laptops at a table, drafting text with AI help

    How to apply it in your business?

    If your team writes with AI, or receives text that might come from an AI, here's what's practical:

    1. If you publish content made with Claude, assume it may carry the mark even if you copy it into another document.
    2. If you review other people's text (candidates, vendors, students), don't decide on a single signal: the mark can be lost through translation, heavy edits, or short text.
    3. Set a clear policy on when and how AI can be used for writing, instead of relying on a detector.

    For another example of reviewing an AI's work with judgment, read the full breakdown of Claude Legal Artifacts.

    Frequently asked questions

    Does Claude watermark its text?

    Yes: Anthropic documented an invisible watermark in Claude's text on November 3, 2026.

    Does copy and paste erase it?

    No: the mark travels inside the text even when you copy it.

    Can I detect it myself?

    For now, detection is only in a private preview.

    Does it prove an AI wrote a text?

    No: it's a clue, not a verdict. Translating, heavy editing, or very short text can make it lose the mark.

    Conclusion

    Claude's invisible watermark is a useful clue, but not proof. Text without the mark doesn't show a person wrote it, and text with the mark shouldn't be used to accuse anyone without more context. Tell us what you write most with AI and we'll tell you if the mark affects you, and send this to someone who writes with AI.