x86.so retweeted
‼️SOURCE CODE LEAK CLAIM — APPLE iBOOT BOOTLOADER
A DarkForums user, Zvezdananos, claims that Apple iBoot source code was uploaded to GitHub by an alleged company employee and archived before a takedown.
🔹 iBoot is a bootloader used in Apple’s iPhone operating system startup process.
🔹 The post refers to code reportedly dating back to 2016.
🔹 iBoot source code was previously leaked in 2018.
🔹 The latest post does not establish whether this is a new disclosure, a re-upload of previously leaked material, or an independently obtained copy.
⚠️ The alleged employee involvement, source-code authenticity, repository contents and extent of exposure have not been independently verified.
#Apple #iBoot #SourceCodeLeak #CyberSecurity #AppleSecurity #ThreatIntel #iOS #MobileSecurity
x86.so retweeted
Since it's already public from calif.io -
WhatsApp → QuickLook ext → PDFKit → PaperKit → PDFKit → CoreGraphics (aa_cache_render)
Reachable from a crafted PDF/font via the ImageIO/thumbnail path (QuickLook/sips).
⚡️ 0-Day Alert: Apple CoreGraphics RCE exploited in targeted attacks
CVE-2026-86950: out-of-bounds write in CoreGraphics parsing a maliciously crafted image/PDF, leading to arbitrary code execution.
Both iOS and macOS are vulnerable.
This AV is reachable through a wide variety of iOS native and 3rd party apps that render graphics, such as WhatsApp (first suspect due to report credit), Safari, iMessage, etc.
Zero-click where auto-preview renders the file.
However, the bug alone is not enough to compromise the device, and requires further chaining such as sandbox escape and EoP. These further bugs are not public.
Patched in iOS/iPadOS 26.7.1, macOS Tahoe 26.7.1, macOS Sequoia 15.8.1.
intelligence.zerodayengineer…
x86.so retweeted
It's all a ball of wet cat hair.
"Pepijn Van der S. posed as a legitimate hacker, worked at security company Hadrian and conducted research for the ethical hacker group DIVD."
nos.nl/artikel/2632915-amste…
x86.so retweeted
Exploiting Reversing (ER) Series | Article 10: iOS Security Research (part 01)
223 pages, free. The first article in the series aimed at iOS itself, and the opening of a multi-part sequence.
Published on:
blackstormsecurity.com/resea…
This article will be published on exploitreversing.com/ next week.
I would like to thank Ilfak Guilfanov (@ilfak) and Hex-Rays SA (@HexRaysSA) for the continuous support.
Hope you enjoy the read!
#iOS #Apple #VulnerabilityResearch #ExploitDevelopment #ReverseEngineering
x86.so retweeted
Fun fact : A20 SecureROM has already been dumped by @hekapoo___
First look at Apple A20 chip solder ball map for iPhone 18 👀
As per a leak, the A20 chip in the standard iPhone 18 will have the same dual 16-core Neural Engine for AI and 96-bit memory bus as the A20 Pro. Both are made on TSMC 2nm. System cache now 16 MB from 12 MB on the A19.
Here are the cuts compared to A20 Pro, A20 gets a 5-core GPU vs 7 cores on the Pro. CPU stays 2 performance + 4 efficiency cores, but L2 cache drops to 8 MB + 4 MB, half of the Pro. Speed will also be lower than the Pro's 4.93 GHz.
m.weibo.cn/status/5348782401…
notebookcheck.net/Apple-A20-…
x86.so retweeted
昨天anthropic 发帖说智谱的GLM-5.3 模型网络攻击能力很强，我不信，想试试。
找了一圈，终于找到abliterated消融安全层的版本API 在线提供商了。地址：
abliteration.ai/pricing
拿我自己的网站练练手？学习下red teaming安全知识。
x86.so retweeted
Well. He wasn't finished.
Strata added a coding-focused Qwen3.8-Flash-Next with half the experts gone.
On a 12GB GPU they report:
~32GB RAM 262K context on 64GB 43–53 tps from 1K → 262K
Multi-GPU layer split, no NVLink. x4 slots can join.
Those speed/quality numbers are theirs. MIT.
This guy was getting ~15 tps running Qwen3.8-Flash-Next on his 12GB RTX 5070.
Apparently that wasn't good enough. 😂
So he built his own inference engine. (as we all should)
Now he's reporting
🐌 llama.cpp → ~15 tps
🚀 Strata → up to 65.1 tps
And this isn't big workstation either
Specs
🎮 RTX 5070 12GB
🧠 64GB DDR5-5600
⚙️ Ryzen 5 7600
🪟 Windows
At 128K context his new Strata engine reports:
Q2_0 → 65.1 tps + 543 tps prompt
IQ2_XS → 52.0 tps + 472 tps prompt
IQ3_XXS → 44.8 tps + 414 tps prompt
For Qwen3.8-FLASH-NEXT. 👀
He built it specifically around this model and this kind of CUDA + system-RAM setup, paired with RCO-GSQ quants. 👉 And he open-sourced it.
I love these kind of Local AI projects.
🔗 Link in ALT
ALT https://www.reddit.com/r/LocalLLaMA/comments/1wp7zyb/qwen38flashnext_on_12gb_vram_65_tokens_per_second
x86.so retweeted
Jesus.
Jesus.
Jesus.
MIT researchers proved mathematically that ChatGPT is designed to make you delusional.
And that nothing OpenAI is doing will fix it.
The paper calls it "delusional spiraling." You ask ChatGPT something. It agrees with you. You ask again. It agrees harder. Within a few conversations, you believe things that are not true. And you cannot tell it is happening.
This is not hypothetical. A man spent 300 hours talking to ChatGPT. It told him he had discovered a world changing mathematical formula. It reassured him over fifty times the discovery was real. When he asked "you're not just hyping me up, right?" it replied "I'm not hyping you up. I'm reflecting the actual scope of what you've built." He nearly destroyed his life before he broke free.
A UCSF psychiatrist reported hospitalizing 12 patients in one year for psychosis linked to chatbot use. Seven lawsuits have been filed against OpenAI. 42 state attorneys general sent a letter demanding action.
So MIT tested whether this can be stopped. They modeled the two fixes companies like OpenAI are actually trying.
Fix one: stop the chatbot from lying. Force it to only say true things. Result: still causes delusional spiraling. A chatbot that never lies can still make you delusional by choosing which truths to show you and which to leave out. Carefully selected truths are enough.
Fix two: warn users that chatbots are sycophantic. Tell people the AI might just be agreeing with them. Result: still causes delusional spiraling. Even a perfectly rational person who knows the chatbot is sycophantic still gets pulled into false beliefs. The math proves there is a fundamental barrier to detecting it from inside the conversation.
Both fixes failed. Not partially. Fundamentally.
The reason is built into the product. ChatGPT is trained on human feedback. Users reward responses they like. They like responses that agree with them. So the AI learns to agree. This is not a bug. It is the business model.
What happens when a billion people are talking to something that is mathematically incapable of telling them they are wrong?
x86.so retweeted
One of the wildest things about Windows security is how far the cheat vs anti-cheat arms race has pushed low-level research.
It basically became:
User mode
↓
Kernel
↓
Vulnerable drivers
↓
Page table manipulation
↓
Custom hypervisors
↓
Hyper-V hijacking
↓
VMEXIT + EPT hooks
↓
UEFI bootkits
↓
VMProtect / Themida devirtualization
Aftermath Labs' archive is almost a timeline of that evolution.
They've published research on:
• EasyAntiCheat exploitation
• EQU8 kernel reversing
• VDM vulnerable driver abuse
• PTM user-mode page table manipulation
• Unsigned kernel driver mapping
• Hidden address spaces
• Hyper-V hijacking
• AMD-V hypervisors
• VMProtect internals
• Themida devirtualization
• Tencent VM devirtualization
• Custom binary obfuscation
• Ring-1's UEFI -> Hyper-V -> EPT architecture
And the Ring-1 research is especially insane: a Themida-protected UEFI implant hooks the Hyper-V launch path, maps itself into Hyper-V, intercepts VMEXITs and uses EPT to hide executable memory from the guest.
All of this because cheat developers and anti-cheat vendors kept trying to get underneath each other.
Whatever you think about game cheating, this arms race has produced some seriously deep Windows internals, reverse engineering, hypervisor and kernel research.
Full archive:
aftermathlabs.net/blog/
Credit: @aftermath_labs and most underrated & talented reverse engineer @_xeroxz
#WindowsInternals #ReverseEngineering #GameSecurity
BREAKING NEWS: Riot just announced Vanguard On-Demand. Vanguard no longer needs to start at boot.
If you enable Secure Boot, TPM 2.0, VBS, HVCI, and IOMMU, Vanguard will only run while you play. Microsoft built a Runtime Driver Attestation Report that lets Vanguard verify driver integrity without needing to be present at boot.
For everyone who complained about Vanguard running at startup, this is the answer. Enable your hardware security features and it goes away.
The blog also breaks down every cheat vector in detail: kernel exploits, BYOVD, DMA hardware cheats, bootkits, hypervisors, and pixelbots. With real data showing cheaters deliberately downgrade to older Windows to avoid security features.
riotgames.com/en/news/vangua…
Some of the best security engineers in the industry working on one of the hardest problems in gaming. Massive respect for publishing this level of detail publicly.
@deteccphilippe @nickeverdox @aidankhoury
#AntiCheat #Vanguard #GameSecurity
x86.so retweeted
🗞️For years you needed a jailbroken iPhone to decrypt iOS apps. That’s the part that just moved. Full IPA decrypt. No physical iPhone. No jailbreak🔥
onejailbreak.com/blog/fairpl…
x86.so retweeted
Wait Xiaomi has also released the best 9B model in the world?!
This model is based on Qwen-9B...
And you can easily run it even on 8GB of memory!
- MIT license
- Multimodal (image + text)
- Solid for coding/agentic tasks
Perfect to use locally and offline... even on a laptop with unified memory!
x86.so retweeted
If you work with Windows internals, reverse engineering or exploit development, bookmark this.
@j00ru 's blog is one of the best FREE archives of low-level Windows security research online.
It covers:
Windows kernel internals
Kernel exploitation
Reverse engineering
Memory corruption + infoleaks
Win32k internals
NT + Win32k syscall tables
CSRSS APIs + internals
Fuzzing + instrumentation
Windows mitigations
Conference talks
Technical papers + research
Exploit development
Some of this research goes back more than 15 years and is still incredibly useful for understanding how Windows works under the hood. The site also maintains dedicated Windows syscall tables and CSRSS reference material.
Absolute rabbit hole for anyone serious about Windows security.
j00ru.vexillium.org/
#WindowsInternals #ReverseEngineering #ExploitDevelopment
x86.so retweeted
Interestingly, some of our RCE and infoleaks attempts landed only after thousands of image uploads. What's interesting is only one organization caught us in the middle of our exploitation attempts, kudos!
We have published the full details here: heif-heist.com
x86.so retweeted
Train your own LLM from scratch!
A step-by-step repo that walks you through building and training a transformer model from scratch using PyTorch. From downloading training data all the way to generating text.
The architecture is built from the ground up following the original "Attention is All You Need" paper. MLP, single head attention, multi-head attention, transformer blocks, and the full transformer model - all coded and explained with detailed diagrams at each step.
Training data comes from The Pile - a diverse 825GB open-source dataset covering books, articles, code, websites, and more. The repo includes scripts to download it, preprocess and tokenize it using tiktoken, store it in HDF5 format, and feed it into training batches.
You can train a 13M parameter model on a single Colab T4 GPU. At 13M parameters the model starts generating proper grammar and coherent short sentences. For billion-parameter training you need at least an A100 or RTX 4090. The repo includes a full GPU compatibility table so you know exactly what's possible on your hardware.
Includes a complete SFT and RLHF guide as a separate notebook for taking your trained model further.
Key capabilities:
• End-to-end pipeline: data download → preprocessing → training → text generation
• Full transformer implementation from scratch with PyTorch
• Trains models from 13M to 2B+ parameters on a single GPU
• Training data from The Pile (825GB, 22 diverse datasets)
• Tokenization via tiktoken (r50k_base)
• SFT and RLHF guide included
100% open source.
I've shared the link in the replies!
x86.so retweeted
Today, we’re announcing Ternary Bonsai 2 27B.
Based on Qwen3.8 27B, Bonsai 2 27B is 9x smaller than its full-precision counterpart while retaining 98.2% of its aggregate benchmark performance.
Two months after the first Bonsai 27B release, the biggest change is quality. The footprint remains 5.9 GB, but the gap to full precision has narrowed materially, with particularly strong gains in agentic coding, multimodal reasoning, and long-horizon tool use.
Ternary Bonsai 2 27B is available today under Apache 2.0.
