Perfectly secure steganography using minimum entropy coupling

I’m a co-author on one recent (cited) work, Meteor. The challenge in LLM/stego is that the model outputs a probability-weighted vector at each step. Ordinarily you just sample a token from this vector using a random number generator. For stego the idea is to sample from that vector using a string of (pseudorandom) input bits that represent your message, in a way that (1) gives you a valid sampling distribution and (2) is recoverable, meaning that given a second copy of the model plus each token it outputs you can recover the bits used to sample. The first part is relatively easy, it’s the second part that’s “hard” in the sense that you either waste a lot of bits (meaning an inefficient covertext) or you end up with sampling that isn’t statistically identical to a random sampling approach. In Meteor we used a practical approach that’s fairly efficient but trades a bit of statistical perfection. This paper focuses more on the theory of how to do this optimally, which is nice! I don’t fully understand the result right now (it’s a tough paper to read) but this kind of theoretical improvement inspired by previous work is nice.

Right, so in total ignorance what I would do is just make a GPT powered text compressor and decompressor using a range coder with its probabilities at each token set by the model. Care would need to be taken with precision and end termination so a bias wouldn't be introduced (more care than typical in compressors, since no one normally cares much about 0.01% inefficiencies-- things like end termination have been addressed by bijective range coders).

Then to use it for stego just take your headerless ciphertext and decompress it. Tada: you get model output. To decode the stego just compress it. Assuming everything was built right and that your ciphertext is uniform, the output should be indistinguishable from the model just sampling using an RNG.

As a bonus, you don't even have a stego tool on your computer you just have a particularly carefully constructed text compressor and decompressor that is perfectly usable (and even state of the art in compression rate, given a big enough model) for the compression application.

This is to transmit confidential information without using encryption. Something I predict will be more and more needed, as democratic governments succumb into their autocratic temptations and attempt to ban encryption.

A higher level of abstraction generative AI will soon be able to apply is in the design of algorithms that humans cannot possibly comprehend that go well beyond rudimentary newspaper want ad spycraft.

Also, image based steg:

https://www.provos.org/p/outguess/

https://www.provos.org/p/detection-with-stegdetect/

The point of steganography (as discussed in the paper) is not unerasable watermarks but undetectable (to the adversary) messages in innocent-looking communication.

I'm confused why you focus on plagiarism detection. That being said, your scenario is very briefly mentioned in the conclusion and requires augmenting the approach (entropy coding) with error correction.

The result would be that as long as your modifications (reordering clauses, etc.) reasonably closely follow a known distribution with limited entropy (which I think it clearly does, although specifying this distribution and dealing with the induced noisy channel might be very hard), there will be a way to do covert communication despite it, though probably only a very small amount of information can be transmitted reliably. For plagiarism detection, you only need a number of bits that scales like -log[your desired false positive rate] so it would seem theoretically possible. Theoretically it also doesn't matter if you use text or images, though in practice increasing the amount of transmitted data should make the task a lot easier. However, I'm not sure if something like this can be practically implemented using existing methods.

This is fairly theoretical work. It assumes that the parties (and the adversary) know the distribution precisely.

Its direct practical may be potentially somewhat limited because people aren't going around communicating randomly selected LLM outputs... and if you use LLM output in a context where text would be expected it could be distinguished.

It's not useful for watermarking as the first change will destroy all the rest of the embedding.

I can make a contrived example where it's directly useful: Imagine you have agents in the field, you could send out LLM generated spam to communicate with them. Everyone expects the spam to be LLM generated, so it's not revealing that its detectable as such. This work discusses how you can make the spam carry secret messages to the agents in a way that is impossible to detect (without the key, of course) even by an attacker that has the exact spam LLM.

Less contrived, a sufficiently short message from a sufficiently advanced LLM is probably indistinguishable from a real post in practice. -- but that's outside of the scope of the paper. It's hard (impossible?) to rigorously analyze that security model because we can't say much about the distribution of "real text" so we can't say how far from it LLM output is. The best models we have of the distribution of real text are these LLMs, if you take them to BE the distribution of real text then that approach is perfectly secure by definition.

But really, even if the situation it solves is too contrived to be a gain over alternatives, the context provides an opportunity to explore the boundary of what is possible.

> I suppose that under authoritarian regimes you will soon have to cryptographically prove that you generated your random bits deterministically from specific keys

I guess that's sarcasm but I don't really get it.

Just in case you mean it, that doesn't seem even remotely technically doable to me. And even if you managed to make people generate all randomness from a fixed PRNG with key escrow, you would have to check what they did with that randomness. If you're willing to go that far, the obvious way to go is to just ban the internet completely (and perhaps also people talking to each other).