HAIJA INTEL REPORT

Finally got around to getting a working PoC of my native object file to LLVM IR lifter. Started this small side-project to see how far can I push my own VM tech and perhaps transpiling existing tradecrafts, executables and post-ex toolings by simply lifting them up to LLVM IR

Ever spotted someone from InfoSec Twitter in the wild and chickened out on saying hi? Yeah, us too. Come embrace the social awkwardness in a safe space where everyone's just as nervous as you are - but also just as excited to connect. @Jhaddix @shehackspurple @Cthulhu_Answers

Teaming up with @GraySwanAI on this one. The premise: don't prompt the model, poison what it reads. Hide your instruction in a tool output or a repo and get the agent to act on it. $30K pool, biggest they've run. Free to enter. No exploit code required, you're writing text. Indir

Happening now! @bagelbyt3s is kicking of his briefing session presenting original research into a new Attack Path technique that results in full WSUS infrastructure takeover. #BHUSA

Based on the research and Def Con talk I did with @_EthicalChaos_ two years ago, with some new techniques in the mix as well.

AI agents don’t pentest like humans, and @KoyalwarTarun ’s BSidesLV research shows just how differently they operate. Across 54 black-box web app targets, the agents often skipped the familiar recon → crawl → enumerate → test workflow. Instead, they recognized likely

Our team is off to a hot start in Vegas @KoyalwarTarun 's BSides research ran AI agents through 54 black-box web app targets, no source code, no hints. The research compares leading open-weight and closed frontier models across real-world security tasks, highlighting where