We started analyzing a classic case where GDB creates the illusion that code in writable memory is corrupted. We ended up discovering two interesting behaviors.
Did you know there's a way to hit a breakpoint without using hardware or software breakpoints? Or how GDB patches the binary to execute an instruction?
Learn more about how GDB works under the hood.
Why is my shellcode being corrupted?
https://allelesecurity.com/why-is-my-shellcode-being-corrupted/
@cynicalsecurity Can't wait to see if LLMs bug hunting is deep or shallow. Really curious. Also interesting how they all turned into bug hunting as their main target and AGI etc talk is long gone already LOL Guess bugs make better PR, financial returns, time will tell :-)
"Feed the fire, let the last cinders burn."
My IB-01: CEL 240 illustration is now available as a print <3
Someone's AI agent has been performing a wide variety of manipulation to the project for a while to the Fedora project. https://lwn.net/SubscriberLink/1077035/c7e7c14fbd60fae9/
It's clearly linked to an account that precedes the, ahem, "agentic AI era", but it also seems the account wwas probably compromised, but everything is unclear, including motivations or the extent of damage.
Blogged about the time I doubled our users by doing proper engineering instead of React slop
@buherator I can verify that the exploit still works with that mpengine.dll version and the 1.453.28.0 definition update that got released 6/10/2026 6:29:20 PM. It takes more than one attempt, sometimes up to 6-7 now, where it used to be almost always one shot, but it still works.
I'm waiting for my other VM to fully update and then I'll retry there, too.