Want to write kernel exploits or analyze malware? You need to understand Windows internals first. Syscalls, IRQL, pools, mitigations & more. Architecture 2821: Windows Kernel Internals 2 by Cedric Halbronn @saidelike https://ost2.fyi/Arch2821

Oh lovely. Our government is investing €500m into slop, and are joining the AI Gigafactory program.

Can the bubble burst already, please?

https://bynar.io/blog/linux-has-screensharing-rces-too
Welp time to update

PLACES/TAXCO7.GIF

How accurate have Ed Zitron's AI skeptic predictions been?

https://danluu.com/zitron/

Interesting Git repos of the week:

Bugs:

* https://github.com/MSNightmare/FalconFlank - everyone's favourite new source of 0day pops CrowdStrike Falcon
* https://github.com/MSNightmare/HardBreacher - everyone's favourite new source of 0day pops Kaspersky AV

Hard hacks:

* https://github.com/salmg/SIMtrace2-T1-Research - research into SIM cards
* https://github.com/i12bp8/TagTinker - messing with IR on the Flipper Zero
* https://github.com/nickdaria/wyze-bulb-color-pwned - and Nick opened the firmware and set "let there be light..."
* https://github.com/jprx/darwin-vm - the Apple never falls far from the tree

Hardening:

* https://github.com/a13xp0p0v/linux-kernel-defence-map - mapping the Linux kernel attack surface
* https://github.com/ryancdotorg/libclaymore - @ryanc defuses dd

Data:

* https://github.com/going-doer/Paper2Code - generate PoC code from papers 🤖
* https://github.com/adulau/ptrclassify - what's in a name asks @adulau

#security, #research, #code

make bad art

What happens if a model is repeatedly asked to make a small, localized tweak to an image. In case you were wondering. You can thank me later.

[ Edit: article with more goodies here: https://blog.coredump.cx/p/recursion-into-madness ]

Yes, we can.

PC Gamer: Go grandmaster becomes first human to win series against advanced AI engine in 3-hour match https://www.pcgamer.com/software/ai/go-grandmaster-becomes-first-human-to-win-series-against-advanced-ai-engine-in-3-hour-match/

Oh my:

https://cert.pl/en/posts/2026/09/vulnerabilities-in-mikrotik-routeros-actively-exploited/

#routeros, #threatintel

The fedi isn't like email.

The fedi is like if you triple booked a BDSM meet up, a FSF convention, and a communist workers AGM all in the same German arena.

\o/ #klutshnik server now also builds with #zig v0.16 \o/

https://rad.ctrlc.hu/nodes/rad.ctrlc.hu/rad:zogkw4qiTrH7rKVa89BrKo8z5miA/commits/699e50ac9301b94b4854438f9021ef73160709f9

was quite easy after updating zphinx* before.

https://klutshnik.info - unlike the common tools like pgp and age - is an online threshold data-at-rest crypto system that provides very cheap and efficient key-updates and thus forward secrecy and post-compromise security.

*zphinx is an online threshold password manager far ahead of the competition in terms of security guarantees: https://sphinx.pm

The opposite of mass-destroying books to feed AI is this 1-minute video 📽️ of Charlotte Wainwright restoring and saving old/ancient books ...

(Definitely worth a watch)

https://www.wainwrightbookbinding.com/

#bookstodon #books

Hell is other people's Makefiles

James Gosling, creator of Java, asked by Navin Reddy whether he codes with AI (somewhat cleaned up from the transcript):

"NR And do you use AI to do that?

JG: I have played with AI tools and I mostly find them to be really annoying. I mean, the thing is that I know how to write software. Um, and every time I've used an AI tool, I I look at the software that it generates and for the most part it's just awful. And I care about how evolvable a piece of software is, how reliable it is, you know, how testable it is. And I have yet to find a coding system that works for me cuz it just all you know it all feels like a really great demo, just not ready for prime time.

NR: So do you think there will be a day when AI can reach there?

JG: Oh maybe, kind of, but why? I mean you know in the spectrum of
problems in the so in software engineering, coding has never really been a big problem. The places where there are problems are in like testing. How do you know that this piece of software actually does what you need it to do? That's a really really difficult
problem. That's a problem that is unpleasant. It burns a lot of a lot of time. And you you know, everything I've seen about AI tools just is that they just make hard problem worse. And the the easy problem, yeah, marginally better. But also the easy problem is also the enjoyable problem. So I view AI in part as stealing the fun and making the torture worse. And you know, people go, "Well, you
know, I wrote up this little Swift iPhone app. Doesn't that count?" And it's like, "I don't care, right? I mean, small toy apps like that are are nothing. I mean, I'm, you know, software doesn't get interesting until it's got like until it's like millions of lines, right? You know, until you're doing something like, you know, operating FedEx, right? Right. You know, and you know, some, you know, a place like FedEx, you know, they've got hundreds of thousands of llines of code, probably actually probably millions of lines of code and, you know, 100 000 engineers. I mean, it's a really complicated thing to run a business like Uber or FedEx or UPS or a bank. And building software suites like that
is is where there's difficulty, you know, glossing up, you know, a Swift app to take a survey. Okay, that's fine. But that's not a problem.

NR: That's right. So when you're building a give it business where you have as you mentioned million lines of code and then we have entire team I mean teams of team who is working together to get the one thing that's where it excels

JG: where what excels?

NR: the human

JG: oh yeah I mean when when it when things get complicated you need people. I have not seen a scrap of evidence that the AI coding tools, you know, if if you wanted to build the
next FedEx and you were just going to vibe code the next FedEx, I think you'd be in trouble.

NR: Yeah. Yeah. You never know when you have security flaw and getting hacked.

JG: Yeah. Yeah. O when when your AI tools decide, you know, the best way to be profitable is to steal money from all
your customers. Because you know then people have tried to to to do
experiments like have AI just run a
single vending machine as a business
or do security penetration experiments
inside inside a company and then you
discover that all hell breaks loose, right? Um these these things are a very,
very long way from being deployable at scale with a decent amount of faith that things will work out."

https://youtu.be/LxhZhOOYKTM?t=513

#AI #Java #SoftwareDevelopment #SoftwareEngineering

CVE ID: CVE-2026-85046
Vendor: Google
Product: Chromium V8
Date Added: 2026-09-04
CVE URL: https://nvd.nist.gov/vuln/detail/CVE-2026-85046