Richard Johnson

defcon 33 main stage sunday aug 10 noon - flyer with sick ascii by x0 and aNACHRONiST

Are you going to @defcon.bsky.social??

We'll be giving away 9500 print copies of Phrack!

Come by main stage Sunday @ noon to see @vacci.ne @richinseattle.bsky.social and chompie talk hacker history! This will mark the first time Phrack staff appear together on DEF CON’s main stage.

AI Agents for Cybersecurity

by Richard Johnson (@richinseattle.bsky.social)

www.hexacon.fr/trainer/john...

Applied Deep Learning AI for Cybersecurity from April 22 to April 23/2025 — secwest.net - secure virtual engagement This class is designed to introduce students to the most effective tools and techniques for applying cutting-edge deep learning–based artificial intelligence to cybersecurity tasks. By leveraging AI-d...

@richinseattle.bsky.social Richard Johnson, at the forefront of AI Security Research for years, is doing a new two day dojo course at CanSecWest2025_newtype, before 24/25 conference.

Check out his presentation last year at CanSecWest "The DL on LLM Code Analysis " at

secwest.net/2024-recordi...

Well the initial vibe check isn’t great but I need a larger sample set. QwenCoder 2.5 did an okay zero shot but since it's not a thinking model I followed up with a prompt of "are there any bugs in the harness?" and it went off the rails producing incoherent output

Hah well the previous benchmark I created was evaluated on a curated vuln identification dataset which made it trivial and allowed me to iterate on the input side which led toward agentic prompt selection. I have other ideas and would incorporate more variables for this task ofc ;)

I was going to include it. Will give it a shot. I need to create a new benchmark to give more empirical measurements. I have a plan for that.

Total awareness, understanding marketability of local models, determining which may be competitive with tuning, etc. Current interest is for which to use for agent dev and tuning for the training I’m developing. Will contrast vs best commercial options as well. Students will have diff applied needs.

Generated a few fuzz harnesses using new local models, OlympicCoder was best, fixing own bugs zero-shot & few hallucinations

Open R1 OlympicCoder 32B
DeepSeek R1 Distill Qwen 32B
QwQ 32B
Gemma-3-27b-it

All 4bit quant. Coder was by bartowski, the rest were Unsloth dynamic quant

My new APPLIED DEEP LEARNING AI FOR CYBERSECURITY training class is now available for sign ups at @reconmtl for 5500 CAD ($3838 USD) early bird pricing

Full syllabus and registration here:
recon.cx/2025/trainin...

My Applied Deep Learning AI for Cybersecurity training will be at RECON’s 20th anniversary con! I have a fuzzing harness gen section but will also cover model training/tuning & AI agents w/ applications in malware, RE, bug hunting, and web app pen-testing. There are also 3 other fuzzing trainings!

Kevin Mitnick Part 01 (Final)

Friends, FBI has responded to my FOIA request for Kevin Mitnick's files, and have made them available to everyone via the FBI public portal here: vault.fbi.gov/kevin-mitnic...

Get your macOS 15.2 xnu CodeQL database here! 🎉

github.com/blacktop/dar...

Visiting family in western Arizona, not too far from Lake Havasu

Mount up, time to ride!

New Year, New Life.

If you say “Pet Cemetary” three times in the mirror.. REDRUM

Happy New Year!

Yep

Nah 3D printed reproduction. I do have an original OKI900 and some red boxes I built in the 90s :)

Support your friendly phreaks! www.etsy.com/listing/1763...

Did anyone else get a blue box for Christmas?

screenshot of the CFP on phrack.org

We updated our CFP for Phrack 72! The deadline is now April 1st 2025. Check the site for specifics on how to contribute, as well as some inspiration! We also posted a link to purchase physical copies of Phrack 71, and a donation link too. Enjoy!

phrack.org

So I guess I'm going to be participating in the Solana hackathon with some LLM magic. I could not have predicted this at the beginning of the year.

News: Murderer was fond of a game where you take the role of an assassin. The game? Among Us. Never mind the cute cartoony aesthetic, this is a training ground for would-be killers! ROFL.

I approve this message. A little digital hygiene will save your ass.

Introducing the Model Context Protocol The Model Context Protocol (MCP) is an open standard for connecting AI assistants to the systems where data lives, including content repositories, business tools, and development environments. Its aim...

If you're developing backend services for the web, you should probably be publishing your APIs using MCP so AIs know how to use them.
www.anthropic.com/news/model-c...

Watch the recording of my #ekoparty talk "Advanced #Fuzzing with #LibAFL" here:

youtu.be/FI7C37lz4Rg?...

Thanks @fede-k.bsky.social for this amazing event!

At that point I should just convert to vector and do my own adjustments. I think that’s what I’m going to try. I dabble in art but I’ve been more focused on the code. Wasted a weekend trying to get GLTF/GLB rendering via OpenGL without a bloated framework, using just GLFW. Failed.

Thanks for your helpful input as always.

I cancelled my midjourney sub. I’ve tried a few options they are all fine at the initial frames but not at making contiguous frames