Sleepwalker: Passive Backdoor with Its Own Command Language | Hacker News Reader

Sleepwalker: Passive Backdoor with Its Own Command Language

r136a1.dev

80 pointsdefrost16 comments

> It is a short sequence of raw bytes that only makes sense when read in a specific order.

This kind of useless sentence is what makes reading AI text so miserable

I have to nope out of half the articles I read here lol

I would be much more surprised by a system of commands which could be read in any order.

I want to make a public blacklist of people who pass off AI writing as their own words. In my view it is tantamount to fraud. It should be disclosed at the top of the page if AI was used.

Agreed, and I've gotten into arguments at work when bringing this up.

I don't think I used the word fraud, to me its plagiarism, but I suppose that's basically a subset of fraud.

The worst part is that the biggest offenders are often management.

I expect X units of information in Y words. Human writing delivers consistent amount of information per unit. I can bail out early. LLM writing ranges between X and 0. Typical addictive variable reward scheme.

Would you not consider this fraudulent?

I suppose I would, which is why I mentioned that while I had called it plagiarism, plagiarism (to me) seems like a subset of fraud. I wasn't trying to get in a discussion whether its fraud or not - either way I don't like it.

How about this one?

> That interpreter is not used just once. When a trigger later delivers a follow-up program, over any of the transports described further down, the exact same interpreter function runs it.

Very interesting, right? An interpreter that is not used just once. Because we all know about those one-time-use interpreters, but this one, is USED MULTIPLE TIMES.

I bet this is Claude, and this is Anthropic’s fix for “it’s not X, it’s Y.”

This is very cool. Definitely seems like either an early or one-time, targeted attack by someone with good know-how and significant resources, even if it itself is not top-tier. Pretty cool find!

in my script kid days, ( 30 years ago ) I did something similar for linux , using raw sockets and I think with ETH_P_ALL. embedding into a an already existing regular linux process ( recompiling the source code ) the problem was high cpu usage of the process when network usage was intensive. of course it did't survive when the OS updated the program.

What would be considered “top-notch” or “top-tier”? Something never found?

Nothing, because blackhat activity disqualifies itself from such a label. Recognition is actively withheld since it'd encourage some bad actors. Same reason malicious software is always given a different name even if one can be found in it.

what..?

Depends on the target, id consider something like Stuxnet to be top-notch given: - Delivery - Target resources (another state) - Target defenses (I like to think that the iranians had good security) - Stealthiness - Target architecture. Its easier to write something for Android (for which we have the OS open sourced) than for siemens PLC

But as others say, you wouldn't want to advertise your exploit as top-notch, as it will bring unwanted attention

This thing has been in the wild since 2024, so it had a good run.

It looks pretty bespoke, so there are likely other variants.

Also targeting the anti-virus executable is a smart way to hide. I don't think this backdoor was cheap to develop.

"Well, you'll never get through the frontline security but you might look for a backdoor."

Reply on news.ycombinator.com