Anthropic publishes technical explainer for the watermark method
4 Aug 14 3:15 PM · 64d ago · 28 articles · 34 posts · 26 comments · 6 sources · development 4 of 8
Anthropic released "How Claude's Text Watermarking Works," describing a steganographic technique that biases token selection at inference time to leave a statistically detectable but individually imperceptible pattern.
Anthropic AI lab, maker of ClaudeJohn Gruber Daring Fireball writerArturo Villarroya Claude Max subscriber, public policy workerVladislav Rajtmajer Freelance developer, Czech RepublicJames Padolsey Author of interactive watermarking explainer
The whole story articlespostscomments the bright band is this development · numbered dots are the others · click one to jump
What was reported 5 claims about this development
-
first by The Independent, 61d ago
-
first by TechCrunch, 63d ago
-
first by Implicator.ai, 63d ago
-
1 outlet How AI text watermarking works
first by HN Frontpage, 65d ago
-
first by Stratechery, 66d ago
What people said 24 voices · best of 44 · verbatim
-
J
Watermarking in LLM generated text. Generate a key, and use it to slightly steer word choices, so that a correlation with the key can be observed when compared against the word choices that would be made with the key. Don't forget that LLMs are autocomplete. https://www. anthropic.com/news/claude-text -watermark
-
Interesting. Here's the section of the EU Act that mandates this:> Providers of AI systems, including general-purpose AI systems, generating synthetic audio, image, video or text content, shall ensure that the outputs of the AI system are marked in a machine-readable format and detectable as artificially generated or manipulated. Providers shall…
-
C
Anthropic has revealed that it will soon watermark content that is processed (not just generated!) by any of its models. In a support article, Anthropic explained that it was rolling out machine-readable watermarks to comply with the European Union’s AI Act, which requires all AI system providers to watermark AI-generated or manipulated audio…
-
I've been working in the media and model IP space for quite many years. What this article misses a bit is the threat model for watermarking in general.Watermarking and fingerprinting have inherently weak security guarantees -- they rely a lot on security through obscurity, weak assumed adversaries to deliver.There are clear trade offs between true…
-
S
Next -> I suppose that we will have to check that our own work doesn't trigger false positives matching LLMs signatures using LLMs 🤯. If intelligence becomes a commodity like electricity, as per Sam Altman's vision, there will be intelligence quotas too (I guess I will never reach mine anyway) "The watermark lives in which words the model picked…
-
How I use claude in my grant writing. I write a rough paragraph. I invoke /concise-mode skill (a supposed instruction that Claude used for their previous concise writing style), and ask it to revise for clarity. I re-read to ensure it says what I wanted, ask for another revision with a specific request, or manually edit.This is a productivity…
-
A
I get C2PA and watermarking images in ways that's harder to redact, but how does this work with text? “''Because the watermark is part of the text, it will travel with the text when it’s copied and pasted elsewhere, and may persist through some editing. Watermarking will be applied at the model level, which means it will be present no matter which…
-
I don't really see how this would work anyway.Even if you find a way to 100% watermark any text, couldn't you just use a non-watermark model? I have a hard time believing every AI company on heart would comply.Hell, even if every AI company on earth decide to somehow apply watermarking to their next model, they would also need to apply it to all…
-
K
Claude aplicará marcas de agua invisibles a texto e imágenes generados por IA Anthropic ha anunciado que los textos e imágenes creados por su modelo Claude incluirán marcas de agua invisibles y metadatos firmados digitalmente. Estas marcas son legibles por máquinas y buscan cumplir con la normativa europea de transparencia en IA. El objetivo es…
-
Would OpenPangram's LoRa fine-tuning training be able to pick up on the patterns of steganography left behind by LLM vendors? https://huggingface.co/pangram/editlens_Llama-3.2-3B(pardon for reposting just now from off-front-page thread on https://declaude.org/watermarking/, just thought of question)
-
A
# EU legislators are demanding that # internet users be informed whether they are engaging with AI and whether the content they see is the output of bots or the creation of a human. --- AI behemoth Anthropic announced that all its future models would support watermarking to comply with the EU rules. https://www. politico.eu/article/artificial…
-
> Google DeepMind tested this impact by serving a model that used watermarking to a portion of their Gemini traffic and comparing thumbs-up and thumbs-down ratings. They found no statistically significant differences from the unwatermarked model. And in a controlled study, human raters comparing watermarked and unwatermarked answers side-by-side…
-
B
Claude watermarking for EU AI Act compliance is getting very close. This is a good summary of what that means. tl;dr: Sub-patterns of largely fungible token choice will skewed in non-distinguishable ways that will, based on volume of impact, indicate likelihood of Claude’s involvement. # ai # anthropic # claude # EUAIAct https://www…
-
This is becoming serious anthropic hasn't even drop the anti detected pattern or mechanism this platform https://founderstoday.org/claude-watermark-remover is claiming to detect it using their own internal pattern but i tried it though it result is quite okay i can't even tell if it worked or not because the text are the same am wondering what…
-
Anthropic has posted a more technical explanation of watermarking . This analogy with Monopoly made me laugh: "On each turn, each player moves a random number of spaces around the board according to the roll of a die. Suppose that, instead of rolling the die to get this randomness, we decided to use a book of the digits of pi."
-
> We will soon be offering a watermark detection API. We’re in the process of working out the details of its implementation.Dumb question - doesn't this defeat the purpose of a watermark? i.e., anyone who wants to avoid detection can simply run `while (has_watermark(text)) text = slightly_rewrite_with_non_anthropic_llm(text)` until it's gone? I…
-
P
User anger over Anthropic’s decision to "watermark" text outputs from the Claude AI chatbot is prompting the company to downplay the potential impact. https://www. pcmag.com/news/facing-backlash -anthropic-explains-why-users-shouldnt-fear-ai-watermarking
-
https://declaude.org/watermarking/ did a good job in explaining how SynthID works. As per their blog, it feels like it will be difficult to remove watermarking on bigger text and the checking for watermarking is also not complex
-
A
This sounds like pattern recognition and then some file data, in the case of having it make a doc instead of just doing the text for one. https:// support.claude.com/en/articles /16266773-how-claude-marks-ai-generated-content
-
It's not just checking user thumbs up/down. As your quote says, they also did a controlled study with people rating the results. What else would you want them to do? The whole point is that it needs to introduce a detectable statistical difference, but humans should not be able to perceive it as a quality difference.
-
B
*Okay, I get it that the AI watermarks are "imperceptible," but what do they look like when they are made perceptible https:// support.claude.com/en/articles /16266773-how-claude-marks-ai-generated-content
-
I wonder how the keys and watermarking survives when you ask the model to write in the style of another author or something that deviates from its classic patterns. I once asked Claude a while back to write some short stories in the Expanse universe and style and be curious if the watermarking survived that text.
-
Anthropic Adds Invisible Watermarks to Claude AI for EU AI Act https://www. geeky-gadgets.com/claude-ai-in visible-watermarks/?utm_source=flipboard&utm_medium=activitypub Posted into AI @ ai-nndflip
-
Would OpenPangram's LoRa fine-tuning training be able to pick up on the patterns of green left behind by LLM vendors?
All 8 developments of Claude's AI text watermark sparks backlash, cancellations… →
Hacker NewsNewswiresMastodonRedditXBlueskyGoogle NewsLobsters