Remix.run Logo
▲ Daniel_Van_Zant 7 hours ago

I'm curious has anyone tried hooking up abliterated large language models to systems like this for physical pen-testing? I know there has been lots of hype lately on the capabilities of large language models at finding and exploiting bugs in online software. I know very little about physical pen-testing but would large language models also open up vulnerabilities in the kinds of things this interfaces with or do digital door locks and such have higher levels of security?

▲monster_truck 6 hours ago | parent | next [-]

I think you're overcomplicating things a little bit. A better example of this would be like what I did recently, where I needed to change some non-user-accessible values on a device, could not get ahold of the firmware, and the spec sheets were all behind NDA walls. Instead, I opened it up and read the flash directly, thankfully it was unencrypted, and let deepseek pick it apart and give me a script to fuck with various values including the one I wanted to edit.

Basically like using a gameshark/game genie with extra steps.

▲app134 6 hours ago | parent | prev | next [-]

I pretty thoroughly reverse engineered the unitree Go2 robot dog with a benchtop mobo from a scrapped bot, a USB-TTL convertor, and Opus 4.6

Used the findings to release a jailbreak so you can program the cheap consumer version ($1.6k) as though it were the expensive research version ($15k)

Very fun exercise

https://github.com/a-bissell/UnLeash-Lite

▲Daniel_Van_Zant 2 minutes ago | parent [-]

Very cool. Super excited about stuff like this getting rid of software-based paywalls on hardware.

▲george_max 4 hours ago | parent | prev | next [-]

I run a startup, InfiShark Tech, where we sell hardware for physical / wireless pentesting.

A new hardware and software product is in the works which will have this type of functionality, including KVM control + AI computer usage. We're aiming to get a Kickstarter out for it early 2027.

▲zf00002 6 hours ago | parent | prev | next [-]

Kind of similar: https://github.com/jayofelony/pwnagotchi

▲mldbk 2 hours ago | parent [-]

Yes and no.

Yes, because technically there are already plenty of solutions out there that can do what the Flipper One does.

No, because Pavel (the guy behind F0 and F1) has explained what makes their approach different, especially when it comes to UX. They even built their own UI kit around their design principles. And what they do basically the level of detailization that nobody goes that deep.

They have done it once with Flipper Zero (many other solutions existed before F0), they do it now again.

▲devmor 7 hours ago | parent | prev | next [-]

I don't know about security devices specifically, but I have semi-successfully reverse engineered some abandoned IoT devices by giving DeepSeek and Claude access to a USB flash writer, logic analyzer and power relay for the devices.

I imagine they could be used to do physical pen testing just as easily. From my experience with digital door locks, their level of security is on average, somewhere between a piece of tape and a $5 padlock.

▲iugtmkbdfil834 7 hours ago | parent [-]

Lmfao, I am tempted to play with that. Did you write up any of those or all just random one-offs?

▲rootsudo 6 hours ago | parent | prev | next [-]

great idea, thank you for this

▲AndrewKemendo 7 hours ago | parent | prev [-]

I have one collecting dust that I stopped using years ago just because I got bored with it so if you wanna work together, I’m happy to. I’d love to see it get more use.

▲exe34 7 hours ago | parent [-]

I think the new one might be better as you can run an agent or server on it and get llm to use it. With the old one you had to push buttons as I understand it?

▲orsorna 6 hours ago | parent [-]

I was just researching and it appears you can communicate with the Zero over usb serial.