Interesting on how Claude watermarking (extending to older models tomorrow) works.
www.anthropic.com/news/claude-...
www.anthropic.com/news/claude-...
How Claude's text watermarking works
Future Claude models will generate text that contains a watermark to comply with the EU AI Act. We answer questions about how it works and whether it affects outputs.
www.anthropic.com
November 2, 2026 at 8:13 AM
Interesting on how Claude watermarking (extending to older models tomorrow) works.
www.anthropic.com/news/claude-...
www.anthropic.com/news/claude-...
Full explanation www.anthropic.com/news/claude-...
How Claude's text watermarking works
Future Claude models will generate text that contains a watermark to comply with the EU AI Act. We answer questions about how it works and whether it affects outputs.
www.anthropic.com
November 1, 2026 at 6:00 PM
Full explanation www.anthropic.com/news/claude-...
AI companies being forced to implement watermarking that only actual writers could circumvent is kind of poetic
How Claude's text watermarking works
Future Claude models will generate text that contains a watermark to comply with the EU AI Act. We answer questions about how it works and whether it affects outputs.
www.anthropic.com
November 2, 2026 at 5:15 AM
AI companies being forced to implement watermarking that only actual writers could circumvent is kind of poetic
👇
oooooooooh
(I'm fascinated by a lot of things but watermarking in texts is definitely in there)
oooooooooh
(I'm fascinated by a lot of things but watermarking in texts is definitely in there)
The watermarks are cleaver, something along these lines (a bit more sophisticated): ieeexplore.ieee.org/document/826...
November 1, 2026 at 5:44 PM
👇
oooooooooh
(I'm fascinated by a lot of things but watermarking in texts is definitely in there)
oooooooooh
(I'm fascinated by a lot of things but watermarking in texts is definitely in there)
to keep in mind re claude’s new watermarking
November 1, 2026 at 5:42 PM
to keep in mind re claude’s new watermarking
Watermarking stuff it generated from literature it stole.
November 1, 2026 at 9:09 PM
Watermarking stuff it generated from literature it stole.
Again, if GenAI’s purpose is NOT primarily deception, watermarking wouldn’t be an issue.
The entire enterprise is built on the premise of deceiving, lying, and fraud.
The entire enterprise is built on the premise of deceiving, lying, and fraud.
Gotta admit I'm getting a lot of schadenfreude from the freakouts about Claude's watermarks and the horrifying idea you might actually have to do your own work to get credit for the work.
November 1, 2026 at 4:56 PM
Again, if GenAI’s purpose is NOT primarily deception, watermarking wouldn’t be an issue.
The entire enterprise is built on the premise of deceiving, lying, and fraud.
The entire enterprise is built on the premise of deceiving, lying, and fraud.
Just ban AI. I'm sure there are ways. I mean Claude is about to start 'watermarking' everything. Somebody, surely, could write something to identify that slop.
November 1, 2026 at 9:23 PM
Just ban AI. I'm sure there are ways. I mean Claude is about to start 'watermarking' everything. Somebody, surely, could write something to identify that slop.
Whatever the method of watermarking, it wouldn't be something you could copy and paste away. You'd have to completely rewrite the text.
November 1, 2026 at 4:51 PM
Whatever the method of watermarking, it wouldn't be something you could copy and paste away. You'd have to completely rewrite the text.
It’s actually SO interesting how they do this and this is the best (interactive!) explanation I’ve seen! declaude.org/watermarking/
November 2, 2026 at 5:25 PM
It’s actually SO interesting how they do this and this is the best (interactive!) explanation I’ve seen! declaude.org/watermarking/
It's really incredibly instructive how mad they get at the watermarking
Gotta admit I'm getting a lot of schadenfreude from the freakouts about Claude's watermarks and the horrifying idea you might actually have to do your own work to get credit for the work.
November 1, 2026 at 10:15 PM
It's really incredibly instructive how mad they get at the watermarking
I am much less categorically hostile to using LLMs for various technical and computational tasks than some people here, but watermarking generative output explicitly can absolutely only be a good thing (although it seems very against their commercial interest)
bsky.app/profile/thel...
bsky.app/profile/thel...
Gotta admit I'm getting a lot of schadenfreude from the freakouts about Claude's watermarks and the horrifying idea you might actually have to do your own work to get credit for the work.
November 1, 2026 at 10:52 PM
I am much less categorically hostile to using LLMs for various technical and computational tasks than some people here, but watermarking generative output explicitly can absolutely only be a good thing (although it seems very against their commercial interest)
bsky.app/profile/thel...
bsky.app/profile/thel...
Ist das dieses LLM watermarking?
November 2, 2026 at 6:27 PM
Ist das dieses LLM watermarking?
I trust that the people who keep leaving prompt text in stuff they present as their own work will be too indifferent to even try to do anything to beat the watermarking.
November 1, 2026 at 5:14 PM
I trust that the people who keep leaving prompt text in stuff they present as their own work will be too indifferent to even try to do anything to beat the watermarking.
This is a really clever method of watermarking AI text - using the "watermark" as the random seed for when Claude has to make a random decision (the example they give is "overcast" vs. "cloudy"), which has no effect on overall word choice but makes it obvious if you know the watermark
November 1, 2026 at 7:11 PM
This is a really clever method of watermarking AI text - using the "watermark" as the random seed for when Claude has to make a random decision (the example they give is "overcast" vs. "cloudy"), which has no effect on overall word choice but makes it obvious if you know the watermark
If you’re referring to the anthropic article on watermarking, they’re saying the pattern of the watermark is undetectable to a reader. As in reading an LLM response with a watermark is indistinguishable from an LLM response without it.
November 1, 2026 at 6:13 PM
If you’re referring to the anthropic article on watermarking, they’re saying the pattern of the watermark is undetectable to a reader. As in reading an LLM response with a watermark is indistinguishable from an LLM response without it.
Thanks! Wow, that's not even watermarking.
November 1, 2026 at 6:07 PM
Thanks! Wow, that's not even watermarking.
Watermarking other people's content/quotes on top of refusing to do the bare minimum alt text is also a choice. Yikes
October 29, 2026 at 10:34 PM
Watermarking other people's content/quotes on top of refusing to do the bare minimum alt text is also a choice. Yikes
Even funnier is that they could still have Claude generate something, then rewrite it in their own words, because the watermarking is just based on statistical probability of choosing certain words. If you change or rearrange the words used, it’s undetectable. They're too lazy to even do that.
November 1, 2026 at 6:24 PM
Even funnier is that they could still have Claude generate something, then rewrite it in their own words, because the watermarking is just based on statistical probability of choosing certain words. If you change or rearrange the words used, it’s undetectable. They're too lazy to even do that.
i see how it is, they're holding off on claude watermarking until after the midterms
Gotta admit I'm getting a lot of schadenfreude from the freakouts about Claude's watermarks and the horrifying idea you might actually have to do your own work to get credit for the work.
November 1, 2026 at 8:45 PM
i see how it is, they're holding off on claude watermarking until after the midterms
I'm glad they're all mad Claude is going to start watermarking.
November 3, 2026 at 8:59 PM
I'm glad they're all mad Claude is going to start watermarking.
this is openai’s website detecting a specific noise pattern watermark in the image. you can run either or both through it. it isn’t just guessing that it might be AI; it is an actual tool detecting one specific thing in the image. you can learn more about the noise watermarking online
November 1, 2026 at 7:02 PM
this is openai’s website detecting a specific noise pattern watermark in the image. you can run either or both through it. it isn’t just guessing that it might be AI; it is an actual tool detecting one specific thing in the image. you can learn more about the noise watermarking online
that's not even their own content, why are they watermarking that shit
October 29, 2026 at 1:36 PM
that's not even their own content, why are they watermarking that shit
The watermarking system is legit more interesting than the actual ai itself.
November 2, 2026 at 3:44 AM
The watermarking system is legit more interesting than the actual ai itself.