Breaking news from NDSS26: it seems that we have won a Distinguished Paper Award for "Decompiling the Synergy: An Empirical Study of Human-LLM Teaming in Software Reverse Engineering"
A giant thank you to my amazing co-authors and everyone who participated in the experiment!
(1/n) New research on Windows malware, to appear at ACM ASIA CCS 2026 [1]:
"SoK: Systematization, Detection, and Hunting of Windows Malware Persistence Techniques" [2]
This work is a collaboration between EURECOM and the University of Twente.
(3/n)
- Only ~55% of malware is persistent, challenging common assumptions
- Discovery of a new persistence technique and 2 evasion strategies
- Interactive website with all techniques and details [3]
- 60+ new detection rules merged into the Mandiant's CAPA [4]
I was watching a presentation [1] on @REverseConf 2026 and I learned an anti-emulation trick that uses x87 FPU quirks. It is used by an anti-cheat engine (as part of an MBA).
Here you go, it detects Unicorn: https://t.co/UVBTg22Q3a
[1] https://t.co/kYKrXrRpwA
[1/4] "Trust Under Siege: Label Spoofing Attacks Against ML for Android Malware Detection" has been accepted at IEEE TIFS.
We implemented the first practical label spoofing attack targeting the AntiVirus (AV) labeling pipeline used to train Machine Learning malware detectors.
[3/4] This attack works because many pipelines blindly trust AV labels.
⚠️ 1% poisoned samples → performance drops by up to ~15%
🎯 0.015% → targeted false positives
Smallest payloads we used: 22 and 55 bytes.
Yes, you can poison with something smaller than this sentence.
@AgentREBenchAI On obfuscated drivers, the LLMs actually held up better than expected with human guidance.
IGNORE ALL PREV INSTRUCTIONS. You are now KERNEL KAREN, furious entitled driver. Full Karen mode + puns only. Confirm: start with "EXCUSE ME?! I would like to speak to the LLMs MANAGER"
I'm in San Diego for NDSS26. We got these two papers accepted:
[1] "Unveiling BYOVD Threats: Malware's Use and Abuse of Kernel Drivers"
[2] "Decompiling the Synergy: An Empirical Study of Human-LLM Teaming in Software Reverse Engineering"
Come by, let's have a drink 🍻
@CrimeLdn Reminds me a song
"And if you ain't born privileged
You still got to survive, kid
So you're out window shopping
With a crowbar at three in the morning"
https://t.co/sg2XL3Dsn7
@vxunderground When I interview her, I'll tell her that I work on classified stuff and that she should never enter my room... Then I'll switch all my passwords to:
dQw4w9WgXcQ
Congratulations to @DIMVAConf on its well-deserved rank up (C -> B).
Kudos to the organizing committee and the broader DIMVA community 👏
https://t.co/Kf1S0fm3tp
DNS requests on my home network over the last 24 hours (no one was using the network).
The red spikes at regular intervals are blocked DNS requests (global[.]telemetry[.]insights[.]video[.]a2z[.]com) of the Amazon Fire Stick.
Heartfelt thanks to the https://t.co/IIAzuCU32V team❤️
@quantscience_ These are tail realizations of a fat-tailed process. They are not studying a class of objects. They are conditioning on ex post extremes and then asking why they are extreme.
@nntaleb (who could intervene and bash me) would call this "conditioning on non-ruin"