Transcript from CTO of GPTZero
https://www.linkedin.com/pulse/how-does-claudes-watermarking-actually-work-alex-cui-v1lee
Anthropic dropped a bombshell announcement yesterday and it was through their help center.
They will be watermarking all text generated by Claude going forwards to comply with EU regulations, in an attempt to trace when text was written by Claude.
Here's what they don't tell you - how does it actually work? As the CTO of GPTZero, I'll share how Anthropic, Gemini and OpenAI are all building text watermarking in this brief deepdive and whether it can be defeated.
Almost all forms of watermarking that are fast and cheap enough for a frontier lab have the same formula, following the KGW method:
In generation:
1. Let's say you've generated n tokens so far. Take those n tokens + a secret key to generate a random hash
2. Use that hash to randomly reweight the probabilities for the n+1 token, and then sample from that new distribution. In the simple case, you could split 50% of all English words into a green or red set based on your hash, and boost the probability of words in the green set.
For watermark detection:
1. For each token, see if it was in the green or red set.
2. To do this, recreate the hash based on the secret key and the text preceding the current token. Then, recreate the green and red set of words.
3. Once you've checked all the words in the text, if the next token is selected disproportionately from the green set more than 50% of the time, you claim the text has the watermark.
I can tell you want to ask the following:
1) Isn't it easy to mess up the hash if you paraphrase the text? The answer is mostly yes, however, you can use a probabilistic function to get your hash instead of a deterministic function (SIR, Adaptive Watermark). Since the entire watermark is a probabilistic model, this is fine.
2) Doesn't this make the text much worse? The answer is yes, it does - Yes, it does – but for most people, it's imperceptible (Google claims in human feedback study with 20,000 texts), since there are exponentially many ways to write the same paragraph. DiPmark does something more sophisticated to avoid shifting the text distribution on average. Of course, watermarks fail on short text or highly predictable texts like "2+2=4".
3) Shouldn't it be easy to figure out the green and red sets? The answer is no. You would need an exponentially large number of samples from the watermarker to reconstruct those sets exactly, but it's a risk if the detector is open to the wild (Watermark Stealing)
Still, there are a couple challenges that a frontier lab needs to overcome:
1. Their watermark needs to work token-by-token because they are streaming their text to users. Many watermark methods plan sentences or paragraphs at a time, or change the text after it's entirely written, in order to make their watermark robust to paraphrasers, and a frontier lab cannot afford to do this yet (SemStamp, PostMark)
2. If the secret key leaks, the watermark is busted. To avoid a large blast damage from this, you need to have a couple secret keys in rotation.
3. There are some texts, like code, that cannot be arbitrarily changed, otherwise the code will break. In those cases, the watermark needs to selectively change words in parts of the text that can tolerate synonyms (i.e. like variable naming) - see SWEET, EWD, Invisible Entropy.
4. They will need to educate their users on how to deal with false positives and false negatives of a detector, which is a big challenge (one we put a lot of effort into at GPTZero)
So, how do I see this playing out in the next 6 months?
1. If Anthropic releases the watermark detector publicly, I think they defeat their own watermark. People find reliable watermark removal strategies by testing against Anthropic (AI detectors like GPTZero have an advantage here because they can train against these adversaries once they become popular).
2. If they keep the detector private to the government, like Google has done, it's "safer". However, there are some papers showing trained approaches that work robustly to zero-shot break watermarks without any data, simply because they try to write the text just like a human (Zhang et al. 2024, Watermarks in the Sand).
3. In my testing, the watermarks don't survive intense paraphrasing (especially if you combine word choice and syntax attacks), or human text substitution (rewrite your AI text by plagiarizing human authors). The free paraphrasers I've tried have quickly bypassed Google Deepmind's watermarking system, SynthID for what it's worth.
4. All-in-all, frontier labs are likely okay with this because they expect most users to not attack the watermark, and also because they + European regulators likely don't care past a certain point - its good enough.
5. Overall, I think users of frontier LLMs will not really care about this, because 1) they don't realize watermarks are there, 2) EU will force everyone to conform, 3) this seems more like regulatory hoop-jumping than an earnest effort from frontier labs to expose LLM use.
For the people who are concerned about watermarks, their first concern should be AI detectors - the systems we are building that are always adapt to new bypassing methods and LLMs.
If you're posting, "its not X, its Y!!", I don't think the watermark is going to make a difference 😁