How many Phrack 73 articles used AI?
— download

A new Phrack issue (73) was released a couple of days ago, yay! Unfortunately, I felt that some article had strong AI-produced vibes, and thus immediately wondered how many authors used AI. OpenAI and Anthorpic have rolled out watermarking for AI-authored content, but verification for text isn't publicly available, so I used Pangram and GPTZero instead, as they seem to be what people are currently using. Of course, those aren't infallible, are bound to make classification mistakes (hence why I used two), prone to false-positives, … but they do provide, in my opinion, an interesting coloration nonetheless.

Here are the raw results:

Article Pangram GPTZero
Introduction 85% Human We are moderately confident this text is entirely human
PHRACK PROPHILE ON Halvar Flake 100% Human We are highly confident this text is entirely human
Linenoise 35% Human We are moderately confident this text is entirely human
Escaping from a Docker container through vDSO 98% Human We are highly confident this text is entirely human
OWNING THE TRUTH 10% Human We are highly confident this text was AI generated
CRYPTOGRAPHIC IDEAS FOR TCC BYPASSES 73% Human We are moderately confident this text is entirely human
Word Machines for Weird Machines 84% Human We are highly confident this text is entirely human
Beyond the Facade :: Exposing Smart Giants 21% Human We are highly confident this text was AI generated
The Art of Staying Hidden 17% Human We are highly confident this text was AI generated
Reflections on Trusting AI - A Virus Inside an LLM 16% Human We are highly confident this text was AI generated
QNX Userland Shellcode Components 99% Human We are highly confident this text is entirely human
UAF Exploitation in the Sheaves Era 87% Human We are highly confident this text is entirely human
From PhrackCTF to Two WinFsp 0-days 92% Human We are highly confident this text is entirely human
fortify-headers 100% Human We are highly confident this text is entirely human
Poor Little Chunks 12% Human We are highly confident this text was AI generated
The Silent Fleet 48% Human We are highly confident this text was AI generated
MOLE - A Binary Ninja Plugin for Static Taint Analysis 59% Human We are uncertain about this document. If we had to classify it, it would likely be considered human
Backdoor attacks against Machine Learning IRL 82% Human We are highly confident this text is entirely human
A Journey Into Kernel Exploit Development in the LLM Era 68% Human We are uncertain about this document. If we had to classify it, it would be considered AI generated
Abusing the Windows Smart Card Minidriver Architecture 100% Human We are highly confident this text is entirely human
Painting Out of the Box 23% Human We are highly confident this text was AI generated
Semantic Abuse of the Linux IIO Subsystem 7% Human We are highly confident this text was AI generated
Disassembly, Flux, and Runtime Hijinks 8% Human We are highly confident this text was AI generated
Mystique - An x86 Metamorphic Compiler 85% Human We are highly confident this text is entirely human
Security is failure: Every Patch Is a Guess 10% Human We are highly confident this text was AI generated
The Moon vs The Shield 100% Human We are highly confident this text is entirely human
The 2000 SSH1 exploit that broke the Internet 100% Human We are highly confident this text is entirely human
The 2003 Linux Kernel Backdoor 97% Human We are highly confident this text is entirely human
BlueBoxing - Myths and Feelings 75% Human We are highly confident this text is a mix of AI and human
The Proxy That Made No Sense 26% Human We are highly confident this text was AI generated

Like ~everyone, I of course do have opinions on this, but I'm more interested here in sharing data than writing yet another opinion piece, the internet already has too many of them.