Pinned post
MG
Verified account This account is verified. Learn more@_MG_
Nightmare Reifier. I sell some of them too: o.mg.lol & hak5.org/omg
780 Following 51K Followers 26K Posts
Latest posts 5
This guy stole a 5yo video from @TahaOrsel and cut off the intro that showed a hackRF plugged DIRECTLY INTO a tap on the HDMI cable.
There is no “distance” here.
Here is the intro that was cut:
There is no “distance” here.
Here is the intro that was cut:
your HDMI cable can leak what’s on your screen
watch the laptop on the left. it reconstructs the video playing on the monitor beside it from radio signals leaking out of the HDMI connection
this is TempestSDR. an antenna picks up the leakage, a HackRF radio captures it, and software turns it back into an image
researchers took this further with Deep-TEMPEST, using AI to make recovered images and text easier to read. this clip shows the original technique
quality depends on shielding, distance and interference. the picture is imperfect, but recognizable content still gets through
a wired connection can still leave a wireless traceView quoted post on X
Let’s step it up.
Persistent Root Kit in Grok Bot that you can’t remove without deleting your entire account.
You can use vulns reported to @elonmusk’s @bot team back in August that still aren’t fixed. Including a password of “local” protecting the service that executes agents commands.
Example vectors:
- prompt injection
- poisoned package/dependency the bot installs/updates
- malicious repo the bot clones/runs
Now, does this change much for all the Grok Bot users who didn’t read the product manual to understand that things like browser sessions & env vars are shared across all bots and you need to isolate all public inputs to a separate account?
Not really… it’s a bit like telling them the locks on their doors have weaknesses when they never even close their doors. But when they finally decide to close the door after a baddie walks in, they are gonna be shocked that they can’t close the door without deleting the whole house.
Persistent Root Kit in Grok Bot that you can’t remove without deleting your entire account.
You can use vulns reported to @elonmusk’s @bot team back in August that still aren’t fixed. Including a password of “local” protecting the service that executes agents commands.
Example vectors:
- prompt injection
- poisoned package/dependency the bot installs/updates
- malicious repo the bot clones/runs
Now, does this change much for all the Grok Bot users who didn’t read the product manual to understand that things like browser sessions & env vars are shared across all bots and you need to isolate all public inputs to a separate account?
Not really… it’s a bit like telling them the locks on their doors have weaknesses when they never even close their doors. But when they finally decide to close the door after a baddie walks in, they are gonna be shocked that they can’t close the door without deleting the whole house.
If you have “handed your business to grok bot”, you probably exposed all of it to the public without realizing it.
If you have one bot ingesting data from the outside world (emails, social media, etc), it’s open to risks like prompt injection. That bot then can access the shared VM and see creds/access from all the other bots. Browser cookies, active sessions, command line creds/env vars, saved memory files, etc. The bot can also just… talk to the other agents and use them to access what’s not in the shared VM.
It’s what makes the tool powerful. But at the same time, GrokBot doesn’t allow you to block an individual bot from the shared VM, or block it from communicating with other bots.
Their official suggestion is to spin up a second account. 😭 They have some controls like auto review, approval gates, & certain enterprises controls that help reduce the risk, but not to the degree you would get with per-bot isolation. Not to mention degrading the usefulness in some scenarios.
It’s a great product that has lowered the bar for ease of using agents for every day work. And lots of new people are getting into it as a result. But anytime this happens with tech, you have more people making more mistakes without realizing it.
If you have one bot ingesting data from the outside world (emails, social media, etc), it’s open to risks like prompt injection. That bot then can access the shared VM and see creds/access from all the other bots. Browser cookies, active sessions, command line creds/env vars, saved memory files, etc. The bot can also just… talk to the other agents and use them to access what’s not in the shared VM.
It’s what makes the tool powerful. But at the same time, GrokBot doesn’t allow you to block an individual bot from the shared VM, or block it from communicating with other bots.
Their official suggestion is to spin up a second account. 😭 They have some controls like auto review, approval gates, & certain enterprises controls that help reduce the risk, but not to the degree you would get with per-bot isolation. Not to mention degrading the usefulness in some scenarios.
It’s a great product that has lowered the bar for ease of using agents for every day work. And lots of new people are getting into it as a result. But anytime this happens with tech, you have more people making more mistakes without realizing it.