Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

How Claude marks AI-generated content

support.claude.com

Post reply on HN
Print view View original

341–350 of 446 posts

Re: How Claude marks AI-generated content

#341

Post by davisr »

godd2 wrote: post #306
matherial wrote: post #299

I guess this is where our true colors show. There's a significant contingent of HNers who always dunk on LLM text detectors and claim that they can't possibly work, that they ruin careers, etc. But now that a lab says "OK, we'll add a real watermark", the reactions are overwhelmingly that it's still somehow wrong. Why do feel so entitled to being able to pass LLM-generated text as our own? I get that a lot of techies…

> Just because we found a "cheat" button doesn't mean it's wrong for others to want to know. One difference perhaps is that you think using LLMs is cheating, while others do not.

Having a ghostwriter in your ordinary life is absolutely cheating. It's letting someone, or something, else write for you and pass it off as your own. Personally, I'm insulted any time someone sends me LLM-generated media.

Re: How Claude marks AI-generated content

#342

Post by kbelder »

Computer0 wrote: post #2

So this won't be happening in the US, but in the EU: " When a supported Claude model generates text, it weaves an imperceptible watermark directly into the text itself. You won’t see it, and it doesn’t change the meaning, quality, or readability of Claude’s response. 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. Wat…

If you do the same prompt with zero noise from the US and the EU, would the difference reveal the watermark?

I suspect they'll roll out the watermark everywhere.

Re: How Claude marks AI-generated content

#343

Post by luxuryballs »

andai wrote: post #330

If I understand correctly, this means that any text with the "watermark" is legally uncopyrightable, including code. Relevant comment from a few days ago: https://news.ycombinator.com/item?id=49203613

what if the output is downstream from copyrightable work? wouldn't the LLM touching it wash that off if this was the metric used?

Re: How Claude marks AI-generated content

#344

Post by tiahura »

andai wrote: post #330

If I understand correctly, this means that any text with the "watermark" is legally uncopyrightable, including code. Relevant comment from a few days ago: https://news.ycombinator.com/item?id=49203613

Not an copyright attorney, but color printers have watermarks. That's never been an obstacle.

Re: How Claude marks AI-generated content

#345

Post by beambot »

Laurel1234 wrote: post #280
beambot wrote: post #39

Yet another reason to support open-weight alternatives, I guess.

Why do you feel the need to deceive readers on whether your content is AI generated?

Not sure I understand your implication... because I'm the primary consumer of the content AI generates for me. I'd rather not have its output adulterated.

Re: How Claude marks AI-generated content

#346

Post by taormina »

New models will mark AI-generated content from day one. Claude models launched in the EU on or after August 2, 2026 will support machine-readable marking at launch. Generated text will carry embedded watermarks, and generated files will include digitally signed provenance metadata where supported.

So, they’ve been doing this for over a week without telling anyone?

Re: How Claude marks AI-generated content

#347

Post by ed_elliott_asc »

Dilettante_ wrote: post #120

I notice the "Limitations" section talks about how content only at some point touched by Claude may return a positive, and content that returns a negative may still be Claude generated. But I really would have liked for them to state explicitly that entirely false positives where a piece is fully human-written may still be marked as generated, because too many institutions with the power to ruin someone's life over t…

It’s worse than that, false positives are possible but someone generating text should be able to get ai to change some words and formatting to break the watermarking, then ai detectors can tell them how well they did.

I don’t know what the answer but I absolutely know it isn’t this.

Re: How Claude marks AI-generated content

#348

Post by qweqwe14 »

Laurel1234 wrote: post #280
beambot wrote: post #39

Yet another reason to support open-weight alternatives, I guess.

Why do you feel the need to deceive readers on whether your content is AI generated?

Why do you feel the need to label the content when the quality of the content will speak for itself?

It doesn't matter where the content comes from, only the quality/usefulness matters. If you are opposed to this idea, the next decades are going to be very tough for you :)

Re: How Claude marks AI-generated content

#349

Post by h0mie »

Laurel1234 wrote: post #280
beambot wrote: post #39

Yet another reason to support open-weight alternatives, I guess.

Why do you feel the need to deceive readers on whether your content is AI generated?

Why is it deception if you never claimed it was unassisted? The assumption now is that most text produced is already assisted by an AI to some extent.

Re: How Claude marks AI-generated content

#350

Post by Groxx »

denverllc wrote: post #326
wrsh07 wrote: post #221

Earlier quoted context omitted.

Scott Aaronson talks about his project at OpenAI to do this^ You can carefully select which pseudorandom number generator (prng) you use to be able to id text of a certain length. I expect there is some performance characteristic you have to manage since you're doing this on every inference, but once you do that it doesn't change the output in any meaningful way (the prng is still a statistically valid prng, it just…

I’ve always wondered how this works when we only observe the final output and not the internal state that’s used to generate the output. The LLM presumably generates f(input, RNG) but we only can observe f(RNG).

Since they do have the input, they could probably just store checksums at each step...

... though I'm not sure why that would be preferable over a coarse rolling checksum over all of the output. Seems like that wouldn't influence output, would be equally imperceptible, and probably easier to calculate (compared to "hash seed times running all LLMs supported times number of RNG algorithms, to see if output matches").

Presumably there's some other trick, or it's a red herring / failed experiment and not what they actually do in practice.

Post reply on HN