Everything I own, owned
53 points by wonk
53 points by wonk
And this is a commercial, guardrailed LLM service. What is precenting us from using it over a cafe/airport wifi to compromise connected devices?
Nothing, you can download abliterated qwen3.8 27b and ask it to hack into anything. Whether it'll be successful is another question, but it's probably the best you can run on average higher end consumer hardware.
Depending on what percentage you want to get, I think just MetaSploit might be enough to cause quite a bit of chaos. Always has been.
Apparently, people who know about Metasploit and Shodan, and know how to use both, do not on average want to burn the critical infrastructure down. I guess LLMs can both help more people find those things, and give more people motivation to see the internet burn, and the question is how large the intersection the intersection will be.
Nothing at all.
Most modern public wifi doesn't allow client devices to talk to each other.
This could also be relatively easy to break in at least the majority of routers we have in the wild.
Got a link? I've never tried... when I wanted to do things like that, a pineapple + a deauth attack was more straightforward (but noisier).
No, not really, I am just assuming that you could point a LLM to the router and try to hack it. Not saying that it is trivial but I assume that at least there should be some routers in the wild that could be abused to give full control or at least some exploit to remove device isolation (it is not like each device is really on its own network in most cases).
A significant fraction of places have device isolation, but if you route through default gateway, you can reach peer clients in the same IP subnet. I used this between my own devices quite a few times.
Wifi configured that doesn't allow devices to talk to each other? I mean it's the cheese model of security, but every layer helps.
If I had the courage to run on some of my more expensive things, I might do this. I wonder how an obliterated model would do?
This is a really cool project and I like it for the demo effect of showing rather than telling.
All of that is a bit scary at first sight, but also likely not something most "smart" devices sufficiently protect against :( This isn't meant to be a dismissive comment, and we should be demanding better security from device manufacturers. But we're not there yet. The article is asking for an idealist better future, and not showing bugs that are exploiting well-known and hardened targets.
All of this "should be" a bug, but I can also see how lots of vendors will say "nah, known issue. Not in scope of what we want to protect against".
I found this comment from HN from someone claiming to have written a new driver really exciting. Assuming it is true (probably a bigger assumption now than in the past), it indicates how great LLM-driven software development could be for end users.
This made me remember that I got a tobii 4c eyetracker sitting in a desk, waiting for the day I find some way to use it under linux.
Months ago I did basically the steps in this blog post on my Tobii Eye Tracker:
thanks!
Software is funny these days. Claude got the eye tracker working in no time at all (including a 'I took a picture but nobody was at the desk, no wonder we see no eyes' comment). But then my window manager of choice doesn't expose the actual position of the windows on the screen, so after a lengthy loop of non-working epicycles , I'm now simulating a pointer.
Yo can somebody do this with the Sonos hardware? I'm so sick of their busted ass integration.
This is also something that I am interested of late. For example, I got interested more interested in Kindle jailbreak scene after jailbreaking both my old Kindle PW3 (with WinterBreak) and my new ColorSoft (with Véra), and I am finding KOReader being a much better experience than the original Kindle software (not necessary from UX perspective since KOReader is infamous for its unfriendly interface, but because I can do pretty much I want instead of being restricted by Amazon's decisions).
I just asked Codex/Antigravity to write a KUAL replacement for my ColorSoft (https://github.com/thiagokokada/kual-next) since lots of interesting hacks for Kindle still depends on its UI and KUAL itself stopped working in recent Kindle firmwares (my PW3 with FW5.12.1.1 works with KUAL, my ColorSoft with FW5.19.6 doesn't).
My next step is probably to actually replace the Amazon software on device. I saw this project here replacing the Kindle FW with an Alpine-based Linux distro: https://forge.solarcene.community/smallsolar/Ponder_installation. Seems like early days, but I may try to get Codex to integrate it with KOReader later since it especifically because this device is really slow with Kindle FW.