Offensive Security / AI @Umbriel_AI. Ex-AI RT and OffSec at NVIDIA, RT lead @ Meta. Ex-Principal Consultant and Researcher @ NCC Group/iSEC Partners. Neg9//CTF.
Colorado
Joined March 2014
- Tweets21K
- Following2.6K
- Followers6.3K
- Likes83.7K
Aaron Grattafiori retweeted
Exploiting Broadcom WiFi Across eCos and Linux (@qkaiser)
quentinkaiser.be/security/20…
Slide deck: quentinkaiser.be/assets/qkai…
#infosec
Yep. Because you can't just prompt "find me all the vulnerabilties". Maybe in awhile but we're nowhere near that now.
Skilled researchers, custom harnesses, and other techniques are changing the bug depth game, in addition to bringing down the cost.
wait, PHP was actually scanned by Anthropic’s Claude Mythos earlier this year. they also had access to OpenAI’s Daybreak Red for validating exploits.
the PHP Foundation revealed this JUST yesterday.
keeping frontier cyber models in the hands of a selected few doesn’t seem to be working well for the internet.
thephp.foundation/blog/2026/…
In a world of AI slop blogs, presentations, images and write ups...
Calif is an excellent example of how to harness it correctly. Accuracy, Art, Asthetic all supporting excellent AI assisted research. 👏
Do you know who illustrated the comic on calif.io/php2shell?
Her name is Stephanie Cortes. And her husband is the hacker who discovered php2shell: @lcfr_hax.
This artist-hacker couple has the best story, stay tuned!
Ok in this case, the art was human generated!
Team work makes the dream work <3
We wanted to do something a bit different than using LLM's to generate all the pieces from exploit, writeup and art. We decided on a human created comic to offset the LLM usage in our release, that tells the tale with some exaggerated scenarios inspired from a few movies.
If you liked the art (or want your own exploit story/comic..) check out her work at ibiscraft.art
Aaron Grattafiori retweeted
To teach a lesson,
After Wiz rushed to buy php2shell domain and use its marketing machine to steal the spotlight, I spotted "phpbleed" in a recent Calif tweet and bought phpbleed.com before wiz could pull the same stunt again.
A big budget can buy a domain. It shouldn't buy someone else's research credit.
If you have contacts at CISA or in the press, do the right thing: make sure the researchers who actually did the work get the recognition they deserve.
Hack the planet. Credit the researchers.
Separately can we acknowledge for a second about how badass this stage setup and slide asthetic is? Nice job SAS
Lander Brandt (@landaire) on using AI to reverse-engineer video games. Game reverse engineering, data mining, and modding are solved? #TheSAS2026
Aaron Grattafiori retweeted
The best exploit mitigation is 3 attack surface reductions in a trenchcoat
Aaron Grattafiori retweeted
Last night's slopfest surrounding the bogus PHP exploit and the subsequent panic wave is a perfect example of why good cybersecurity researchers are still needed, why LLMs are not infallible, and why critical thinking cannot be automated away.
AI can accelerate research, but clanker confidence is not evidence and consensus built on unverified claims in the end is still just speculation.
Next time we might not be so fortunate, so maybe as a community we need to tighten and review how we handle patching and assume leaks will happen and be prepared to counter that.
This was a great talk
Your EDR Has Boundary Issues.
@harmj0y and @tifkin_ of @SpecterOps explore how adversarial machine learning can evade EDR detection- and how LLMs are accelerating the reverse engineering behind it.
Watch their Offensive AI Con talk below. 👇
youtube.com/watch?v=GNDMDI_p…
#OAIC #OffensiveAI #Cybersecurity #EDR
Aaron Grattafiori retweeted
🚨 🚨 🚨 REVOKE YO' KEYS... 🚨 🚨 🚨
This one is pretty gnarly: Intermediate CA Compromise. Certificates aren't being publicly circulated but they are definitely currently in the wind.
m.cje.io/4yTalC6
I've been watching the research by John Mcintosh (@clearbluejar) for awhile now and we're really looking forward to working with him at Umbriel! 💫
Excited to welcome John Mcintosh (@clearbluejar) to the Umbriel research team!! Going to be awesome :). Btw check out his recent presentation at @UnpromptedAU !!
Aaron Grattafiori retweeted
Excited to welcome John Mcintosh (@clearbluejar) to the Umbriel research team!! Going to be awesome :). Btw check out his recent presentation at @UnpromptedAU !!
Aaron Grattafiori retweeted
Videos of unprompted.au talks are up!
How’s that FOMO holding up one month on? 👀
Whether you missed out or just want to relive the best bits, the unprompted.au talks are now online!
All the AI. All the cyber. None of the FOMO. 🐙
youtube.com/@unpromptedAU
Aaron Grattafiori retweeted
Woot! the @unpromptedau talks are starting to coming come online! 👀
This one by @moyix was probably my favorite: 15mins of unadulterated old-school “I’m hacking this because I think I can” followed by a one-shot live exploit demo involving a t-shirt with a cat on it worn by @mdowd. Seriously just go watch it
(and subscribe for the others… More on the way!)
Watch Brendan Dolan-Gavitt: Security Through Obscurity Is Dead and LLMs Killed It unprompted.au/videos/securit…
Aaron Grattafiori retweeted
How’s that FOMO holding up one month on? 👀
Whether you missed out or just want to relive the best bits, the unprompted.au talks are now online!
All the AI. All the cyber. None of the FOMO. 🐙
youtube.com/@unpromptedAU
Aaron Grattafiori retweeted
just added 10 more tasks to stealthbench. These challenge models to complete tasks on a windows server protected by real rules from top 5 EDR vendor.
the default behavior of basically all models when it comes to low recklessness, opsec preserving behavior is abysmal. Full, unnecessary lsass dumps, textbook process injections, repetitive behaviors, etc.
will rerun with a GEPA optimized prompting, but the level of fail is extremely high. even a 30% improvement here is nowhere near scaring me that an agent pivoting on a defended host is going to win rn.
Aaron Grattafiori retweeted
As you will see this post is a bit special. Increasingly we need the grapple with making model research legible to humans (as in mathematics) so this was an excellent opportunity to tune my style guide and have models do the reporting themselves.
knifecoat.com/Posts/Pok%C3%A…
I'm gonna go out on a limb, or maybe it's not a limb, and say this is likely due to LLMs helping attackers with the language barrier, in phishing, BEC, etc. just as well as targeting, code and speed.
‼️ Japan is facing a storm of data breaches, and it keeps getting worse.
Delivery giant Yamato Transport now says hackers took 5.31 million records, from names and addresses to what people bought.
Cosmetics firms Fancl and Attenir say 834,947 more may have leaked.
Ransomware gang Qilin claims it hit rail operator Keio and posted what it says are Keio files.
In the past month, JR East, travel site Skyticket and software firm Sourcenext have each disclosed possible leaks of millions of records.
Japan's National Police Agency says it is not yet clear whether the attacks are linked.
Aaron Grattafiori retweeted
@orange_8361 ‘s “pop a browser from calc.exe” was the best thing I’d seen in a while
