Yet another Hacker News

Agentic backdoors via pretraining poisoning

63 points gmays 3 hours ago 14 comments

geuis 2 hours ago | parent

I've been thinking about this for about a year. In my head I've called it "context bombs", though that may not be super accurate.

Think back to the movie The Manchurian Candidate. Produce highly capable LLM models then give them away for free. Train them with key phrases/codes etc to perform nefarious actions later when those codes are present in some piece of ingested context.

There were a couple of papers written on the topic in 2018 (I think, I'll need to find them again) and there's been surprisingly little more work done on the topic since. At least from what I've seen.

I really love the use of local models over subscription services but personally I would never recommend using open weights models from certain sources in a commercial environment.

sowhat1 2 hours ago | parent

If you were a malicious actor do you really think this method is going to offer the greatest return on investment?

geuis 2 hours ago | parent

Sew a million seeds, see what grows.

jack_h 2 hours ago | parent

Nation states aren’t looking for a return on investment rather they have a particular objective they want to achieve, e.g. stuxnet or the pager attack.

usernomdeguerre 1 hour ago | parent

I think this is a reasonable question. The moment this activates and exfiltrates or destroys a system the entire model family, training pipeline, and company are burned, no? How can you construct plausible deniability if your model will (presumably deterministically) activate on trigger?

c0nsumer 2 hours ago | parent

I just sort of presumed this was already being done. Because why wouldn't it be?

Aperocky 18 minutes ago | parent

Well, if it is being done, per the other article on the front page, we should be able to find it.

gostsamo 2 hours ago | parent

few places up on front page is a post regarding big model improving a smaller one. Lo and behold, next year big models will search for backdoors in smaller models who will be used to get the bigger models out of work.

yodon 2 hours ago | parent

If LLMs can edit weights directly to achieve desired ends as recently reported[0], then this class of Agentic Backdoors likely becomes even easier and more prevalent.

[0]https://news.ycombinator.com/item?id=50341174

yodon 2 hours ago | parent

Downloading models off huggingface feels like the new warez.

Sure, the person who abliterated/edited that model might have given you an amazing unlock for free, or they might have given you a credential stealer.

_pdp_ 2 hours ago | parent

Having strong sandboxing and observability around the models is clearly a hard requirement if you want to run these things autonomously.

0xc0c0c0 1 hour ago | parent

Of course this is authored by people at Anthropic. Continuing their narrative that open sourced models are dangerous and that people should trust the closed sourced providers.

This of course is not an issue if you actually care about security where you would've already sandboxed your agent and put guardrails on the inputs/outputs of the model.

tedivm 1 hour ago | parent

> Poisoning only 0.1% of the pretraining corpus

Considering the scale of the datasets used by most companies, being able to poison one out of every thousand documents used in it is a huge leap.

Then, on top of that, you have to also then get your trigger somehow sent to an agent that will honor it.

I guess I'm just not impressed? Like yeah, if you train a model to do bad things it will do bad things, that shouldn't be shocking to people.

ryan_glass 1 hour ago | parent

Is this only an issue if you are using agents? Or are there any theoretical risks of serving a model over say llama server (llama.cpp) and accessing it for chat in browser on machines on your local network?