• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI
    AI

    Anthropic Rolls Out Claude Text Watermarking

    Anthropic published an FAQ on adding detectable signals to Claude outputs.

    AN
    AJ
    SS
    9 Sources, 49d ago, first seen 49d ago

    TLDR

    Anthropic announced via FAQ that future Claude models will embed watermarks using a secret key. The company states this meets EU AI Act rules, matches steps other developers have agreed to take, and leaves output quality and content unchanged. Andy Jones at Anthropic noted the approach draws from Google's SynthID work. Steven Sinofsky highlighted tracking risks from stored hashes. Nick Dobos replied that wording changes contradict the no-impact claim. The public posts confirm the rollout plan and these reactions.

    Combined views

    11M

    9 Sources, first seen 49d ago

    Combined views

    11M

    9 Sources, first seen 49d ago

    6.7K likes
    6.7K likes
    2K comments
    3.5K saves
    760 reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    2K comments
    3.5K saves
    760 reposts
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    9 Sources

    @stevesiThe key (ha) is that they can store all the resulting hashes with your account even. The issue is not detection, text manipulation, or even the existence of a watermark. The issue is tracking.
    @AnthropicAIWe’ve written an FAQ to answer some of the questions we've received about watermarking. In summary: • We’re implementing watermarking to comply with the EU AI Act. Other major model developers have signed the same Code of Practice and will also be implementing watermarking; • Our watermarking method doesn’t have any practical impact on the quality or content of Claude’s outputs; • The difference between watermarked and un-watermarked text will not be distinguishable to readers; • Nothing is added to the text and there are no hidden characters; • Watermarking doesn’t require extra tokens, and will not be more expensive; • Watermarks can’t be traced to a specific person, organization, or chat. Read more: https://www.anthropic.com/news/claude-text-watermark
    @NickADobos@AnthropicAI @alexalbert__ “Our watermarking method doesn’t have any practical impact on the quality or content” This is a lie. You literally say you change the wording Anyone with common sense about how psychology, writing, and legal documents work, knows this is completely false
    @andy_l_jonescryptographic watermarking chat is extremely technically cool. i first read about it as an aside on Scott Aaronson's blog in 2023 - while he was at OAI - and while we've got a high-level writeup below, there's also really good lecture from Scott here https://www.youtube.com/watch?v=2Kx9jbSMZqA
    @max_spero_Pangram mentioned!
    @TuhinChakrRT @max_spero_: Pangram mentioned!
    @mykolaThis sure seems to alleviate any concerns I heard about privacy or tracking. It’s literally just “it will be possible to know with high certainty when a unit of text was authored by Claude.” Lots to be angry about in this world but this really, really isn’t it. The only reason I can see being angry here is that you object to having a way to tell if text was generated by Claude. Why is that bothering people?

    9 Sources

    @stevesiThe key (ha) is that they can store all the resulting hashes with your account even. The issue is not detection, text manipulation, or even the existence of a watermark. The issue is tracking.
    @AnthropicAIWe’ve written an FAQ to answer some of the questions we've received about watermarking. In summary: • We’re implementing watermarking to comply with the EU AI Act. Other major model developers have signed the same Code of Practice and will also be implementing watermarking; • Our watermarking method doesn’t have any practical impact on the quality or content of Claude’s outputs; • The difference between watermarked and un-watermarked text will not be distinguishable to readers; • Nothing is added to the text and there are no hidden characters; • Watermarking doesn’t require extra tokens, and will not be more expensive; • Watermarks can’t be traced to a specific person, organization, or chat. Read more: https://www.anthropic.com/news/claude-text-watermark
    @NickADobos@AnthropicAI @alexalbert__ “Our watermarking method doesn’t have any practical impact on the quality or content” This is a lie. You literally say you change the wording Anyone with common sense about how psychology, writing, and legal documents work, knows this is completely false
    @andy_l_jonescryptographic watermarking chat is extremely technically cool. i first read about it as an aside on Scott Aaronson's blog in 2023 - while he was at OAI - and while we've got a high-level writeup below, there's also really good lecture from Scott here https://www.youtube.com/watch?v=2Kx9jbSMZqA
    @max_spero_Pangram mentioned!
    @TuhinChakrRT @max_spero_: Pangram mentioned!
    @mykolaThis sure seems to alleviate any concerns I heard about privacy or tracking. It’s literally just “it will be possible to know with high certainty when a unit of text was authored by Claude.” Lots to be angry about in this world but this really, really isn’t it. The only reason I can see being angry here is that you object to having a way to tell if text was generated by Claude. Why is that bothering people?