DongHa Lee

Vulnerability Researcher in KR 👋 /  / [email protected]

[Hyungyu Seo @wh1te4ever]

Had some fun with new usbliter8 bootrom exploit. Got root shell yesterday by booting custom iOS 27.0b2 firmware on iPhone 11 Pro.

Honestly, it's not practical for daily use. For some reason, baseband, Wi-Fi, and USB connections are completely broken.

[Miguel Ángel Durán @midudev]

NVIDIA da acceso gratis por API a modelos de IA.
A DeepSeek V4, MiniMax M2.7, GLM 5.1, Gemma 4...

Perfecto para pruebas de desarrollo.
Puedes usarlo con OpenCode y OpenClaw

Puedes conseguir acceso aquí:
→ https://t.co/4F510HFfFi

[ENKI WhiteHat @ENKI_official_X]

[1/2]
CVE-2026-32223: heap overflow in usbprint.sys (IOCTL 0x220064).

Malformed USB descriptor, Named Pipe spray + Ghost Chunk for kernel leak, forged IRP, SYSTEM.

Writeup: https://t.co/EMSyWCPsVL

#CVE_2026_32223 #WindowsKernel #LPE #vulnresearch

[DongHa Lee @gap_dev]

This is exactly what I needed
https://t.co/tDumtpqaJ2
https://t.co/dL1wBv8LF8

[DongHa Lee @gap_dev]

IPv6 isn’t the end
but this IPv8 draft feels like it was LLM-generated and tries to cram everything in, and bringing in JWT makes no sense
https://t.co/FaSCe77ZiL

[Md Ismail Šojal 🕷️ @0x0SojalSec]

What if I told you the first GPU neural network implementation happened in Korea in 2004? 😗
Before CUDA, Before modern deep learning hype.

The world’s FIRST GPU-accelerated neural network wasn’t built in Silicon Valley or by NVIDIA,

It came from South Korea’s Soongsil University in 2004, years before CUDA even existed.

two Korean researchers, Kyoung-Su Oh and Keechul Jung at Soongsil University quietly implemented neural networks on a consumer ATI Radeon GPU and achieved massive speedups.

Used an ATI Radeon 9700 Pro to get 20-30x speedups on matrix multiplications and sigmoid functions for real-time text detection.

translate then read
- https://t.co/nQQM8W8FYO

[DongHa Lee @gap_dev]

Wrapped up BOB14
grabbed some bounty on the way. Not bad at all.

[DongHa Lee @gap_dev]

🍎 CVE-2026-28817
@G1uN4sh, with @4ncientH @leegn4a @DQPC_lover @R4mbb1 @gap_dev
Discovered by BoB 0xb6 team.

[DongHa Lee @gap_dev]

🍎CVE-2026-28868 XNU :)

[DongHa Lee @gap_dev]

🍎CVE-2026-20695 XNU bug!
[ZDI-CAN-28499]

[DongHa Lee @gap_dev]

@wh1te4ever Great work👍

[Hyungyu Seo @wh1te4ever]

Some demo about running Virtual iPhone using Apple Virtualization framework:
https://t.co/be3rG3zZSE

[DongHa Lee @gap_dev]

@MrKaLi176442 Varies case by case.
Probably tied to the quarter cycle, so from my experience it can take up to ~6 months.
I don’t know exactly though.😭

[DongHa Lee @gap_dev]

Received $1,000 from Apple’s bug bounty program!
Old report, but a nice surprise🎁

[DongHa Lee @gap_dev]

@MrKaLi176442 I reported this bug in June, received a CVE in October, and the bounty has now been addressed.

[DongHa Lee @gap_dev]

@tom_doerr is it stable?

[DongHa Lee @gap_dev]

Getting ready to try something academic🎓
https://t.co/9hKjaH8FRH

[SinSinology @SinSinology]

🤌🔥 "Build a Fake Phone, Find Real Bugs
Qualcomm GPU Emulation and Fuzzing with LibAFL QEMU"
https://t.co/6cYo6nj32r

[Ivan Fioravanti ᯅ @ivanfioravanti]

On Apple Silicon, the GPU by default cannot use more than about 75% of the system's total memory, but you can change this setting easily. I use my own script for that and set it up to 98% on M3 Ultra 512GB.
https://t.co/Oo9XOTHUsd

[DongHa Lee @gap_dev]

I’m not affiliated with Talos.
I’m an independent security researcher, so I don’t really know the details of the other bug.

For my bug, I found it quite a while ago, and it looks like the credit process got missed after I reported it (it’s being sorted out now lol).

As for exploitation: at least when I analyzed it back then, I didn’t see a viable way to get a shell. Even though it’s a stack overflow, I couldn’t find a memory layout that I could realistically overwrite in a useful way.

[DongHa Lee @gap_dev]

CVE-2025-23339
My first NVIDIA CVE🎉 (found via grammar-based fuzzing)