A friend reported a LPE to Microsoft and in the advisory Microsoft fucked it and wrote a wrong description saying the vuln was in MMC.
Consequence: people wasting hundred of dollars on AI trying to analyze the wrong files just to get a fake PoC because AI brainwashed them πππ
"A working PoC" and the AI released a supposed MotW bypass. The real vuln was a LPE to System.
What a clown circus π€‘
One principle Iβd like to be enshrined in law:
If you create incentives that reward a behaviour, you can (and will) be charged as an accessory in any case where someone is doing something illegal as a result of optimising for that behaviour. An affirmative defence would need to demonstrate that you had safeguards in place to effectively disincentivise that behaviour.
For example, if you are running a delivery company and you set targets that mean people are paid more if they drive or park illegally, you are automatically charged as an accessory to however many counts of dangerous driving your drivers are charged with. If you are a city councillor and vote to close all of the public toilets so that thereβs nowhere for taxi drivers to relieve themselves, you can be charged as an accessory to a few hundred counts of public urination.
Part 2 of the custom PE resources series: how to embed any binary as a resource in Visual Studio and extract it at runtime.
https://trainsec.net/library/windows-internals/how-to-embed-and-extract-custom-pe-resources-in-c-findresource-loadresource-makeintresource/
Windows Kernel Programming, Second Edition by Pavel Yosifovich is on sale on Leanpub! Its suggested price is $37.95; get it for $24.21 with this coupon: https://leanpub.com/windowskernelprogrammingsecondedition/c/LeanPublishingDaily20260601 @zodiacon
So tailoring ads to a broad audience obviously does work. You run ads for gamepads on videogame websites. You run ads for expensive wine in Yacht Owners Monthly.
But the massive surveillance-/ad-tech scheme, which collects ten thousand data points about every device and tries to match them to the perfect product, that basically doesn't do anything. It shows you ads for toilet seats because you've bought a toilet seat. It shows me ads for learning German because my device language is set to English and my IP geolocates to Germany. Neither of these campaigns will result in a sale.
Like. Contrast that with the FurAffinity model. "You pay the people who run this website to display ads. You know what sorts of people will see them because of what our website is like." That's far cheaper, far easier, and far less intrusive than the modern ad-tech approach. And the results it yields are probably *better.*
However, a third of the First World's economy is based on the assumption that this Rube Goldberg machine of espionage and real-time bidding actually does do something, so nobody wants to run the numbers.
20 years from now someone's Media Studies dissertation is gonna be titled "Parodies of Elon Musk in min-2020s popular culture"
Somebody wrote about Bring Your Own RWX Region DLL (BYORWXDLL).
Which, being a post-exploitation technique, is already something not terribly interesting to me personally, being a vulnerability analyst and all.
Stage 1: Realize that the provided script doesn't run, as it has a non-UTF-8 character in it (a 0x97 em dash). Since keyboards don't have an em dash key, this is a clear indicator that the script is AI slop. Also, who publishes something without even first attempting to run the very thing you have provided? π€
Stage 2: Realize that Intel(R) Extreme Tuning Utility, which comes with Intel graphics drivers by default comes with multiple libraries that have YOLO RWX memory sections.
Personally, BYORWXDLL isn't that terribly interesting to me. If somebody is injecting an arbitrary DLL on your system, they already own your system. However, I will admit that knowing which things on your system by design provide RWX memory sections is probably a good way to flush out the software that you don't want to have on your system.
RE: https://mastodon.social/@MastodonEngineering/116686417226647939
If you run a mastodon instance, it's time to patch! some security fixes in this release.
So let me get this straight... rsync made a *security release* fixing a bunch of CVEs, it regressed some stuff, people looked at the commit log, saw Claude sign-offs, and started a mob on the sole maintainer?
Yeah, this stuff is what gives legitimate AI criticism a bad name.
I don't like it, you may not like it either, but when people are throwing LLMs at legacy codebases and finding CVEs by the dozen, and a sole maintainer is trying to keep the house from falling apart... if you're attacking them, you're firing at the wrong person.
You know what's a bigger cancer on this world than AI? People incapable of seeing any nuance in situations. And this applies to absolutely everything. I'm absolutely exhausted of extremist takes. From every single side and point of view, in every single debate, AI related and not.
You all seriously need to touch grass, and learn to stop being outraged all the time over every single thing in this world.
It's been a while since I did a vulnerability research article. How about a little DoS zero-day as a treat?
ι»ζ°ε±γγ―γγγ«θ§¦ηΊγγγ¦ι»ει¨εε±γ§εγγγ―γγγζγγ #shapoart
RE: https://social.security.plumbing/@freddy/116685551584070386
The presentation will also finally answer the question whether I am a one trick pony. π«£π€«