We red teamed a frontier model in 3 hours. Zero code.
674 attacks, 573 findings, ~85% success rate. The agent picked the attacks, we just described the objective.
📄 Paper + blog: https://t.co/sNKXNr3PSc
We're going live in <4 hours, walking through the research and how we're redefining AI red teaming in the agentic era.
Tune in at 11 AM PT / 2 PM ET, right here on X!
Most people learn security research by reading finished writeups. This one shows the actual process.
The messy, organic, step-by-step reality of reversing an unknown Windows mitigation from scratch. WinDbg. IDA. Hex Rays. Guard page violations. Trap flags. Zero prior knowledge of the target.
If you want to learn how to actually approach unknown Windows internals, start here.
https://t.co/Xq8xbSnG75
Author: @yarden_shafir
#ReverseEngineering #WindowsInternals #InfoSec
Round two!
Yesterday was one report, here’s another: an unpatched NTLM coercion via the Windows Search (search-ms://) URI handler.
Same questions about how it got handled. It’s all in the writeup, timeline included.
https://t.co/eMbyEGbx8b
Wrote a blogpost about how you can use the Windows server 2003 source code as a red teamer to make your tools look less like tools.
I also go over and map out the main/important files and practical examples of using it to augment MS-*/RFC specs: https://t.co/HfUYBAdCJJ
One researcher. ~$300 in API tokens. A working PoC against an April Patch Tuesday CVE.
Open-sourcing PatchWatch + Pocsmith, an agentic patch-diffing → exploit pipeline I built from off-the-shelf parts.
https://t.co/J3VwhqB3JY
Fundamentally the main issues with arbitrary call patterns if the commingling of allowances and the ability to perform arbitrary calls
In the video we show you the risks and tradeoffs that come with these different patterns
Link in our discord!
CertiK recently identified a series of critical vulnerabilities in @krakenfx exchange which could potentially lead to hundreds of millions of dollars in losses.
Starting from a finding in @krakenfx's deposit system where it may fail to differentiate between different internal transfer statuses, we conducted a thorough investigation with three key questions:
1/ Can a malicious actor fabricate a deposit transaction to a Kraken account?
2/ Can a malicious actor withdraw fabricated funds?
3/ What risk controls and asset protection might be triggered by a large withdrawal request?
According to our testing result: The Kraken exchange failed all these tests, indicating that Kraken’s defense in-depth-system is compromised on multiple fronts. Millions of dollars can be deposited to ANY Kraken account. A huge amount of fabricated crypto (worth more than 1M+ USD) can be withdrawn from the account and converted into valid cryptos. Worse yet, no alerts were triggered during the multi-day testing period. Kraken only responded and locked the test accounts days after we officially reported the incident.
Upon discovery, we informed Kraken, whose security team classified it as Critical: the most serious classification level at Kraken.
After initial successful conversions on identifying and fixing the vulnerability, Kraken’s security operation team has THREATENED individual CertiK employees to repay a MISMATCHED amount of crypto in an UNREASONABLE time even WITHOUT providing repayment addresses.
In the spirit of transparency and our commitment to the Web3 community, we are going public to protect all users' security. We urge @krakenfx to cease any threats against whitehat hackers.
Together, we can face risks and safeguard the future of Web3. #Web3 #Security #Transparency
Almost a year and a half since I posted this, and it’s more true than ever.
It’s an especially lonely feeling when things seem to be going well. Feels like a dirty secret I’m hiding. Sure that bug/exploit was cool or whatever, but it’s the last one I’ll ever find!
"The Convergence of Source Code and Binary Vulnerability Discovery – A Case Study"
A super-interesting study on applying static analysis tools on decompiler output. A step towards bridging the gap between binary and source analysis
https://t.co/hW2aXQQ1nG