øk (@x41ok) header
øk (@x41ok) profile picture

founder of 0x41 · open-source AI security tools · monomi · akashi · kaidoku · trust nothing. verify all.

Joined November 2019
Photo by øk (@x41ok)Photo by øk (@x41ok)Photo by øk (@x41ok)Photo by øk (@x41ok)Photo by øk (@x41ok)Photo by øk (@x41ok)Photo by øk (@x41ok)Photo by øk (@x41ok)Photo by øk (@x41ok)Photo by øk (@x41ok)
Pinned Tweet
Meet 0x41. We build open-source AI tools that turn raw security data into answers you can act on. Two tools so far 🧵 Monomi: give it a domain and it maps subdomains, hosting, open ports, web tech, and security headers. It’s passive by default, and the AI report cites the evidence behind every finding, so you’re never guessing where a claim came from. Kaidoku: give it CVEs and it ranks them by urgency, flags ones tied to ransomware campaigns (via CISA KEV), and explains real-world impact in plain English, tailored to your stack. They work together: feed Monomi’s output into Kaidoku and every CVE keeps the host it was found on. Both are MIT licensed on GitHub. 0x41.sh
6
315
If Iran does attack us and we go into a nationwide blackout please don’t fill your bathtubs with water that water is for Claude
132
2,095
31
50,003
944,718
Claude, delete the IRS. Make no mistakes.
52
284
11
5,448
87,432
1. big night for 0x41. you can now hand claude a domain and it runs the whole chain. monomi finds, akashi proves, kaidoku explains, then it walks you through the fixes. 2. the part i cared about most was active checks. claude can ask to run one but it can't approve it. i run `0x41 approve` in my own terminal, see every request it would send, and type yes. one run, 15 minutes, then the approval is gone. 3. i tried MCP elicitation first (the confirmation popup). built a small test only a real person could pass. the app doesn't support it yet, and some clients have been reported to auto accept without asking anyone. so the yes stays in the terminal. 4. api keys stay out of the claude config, and the audit log ties every request back to the approval that allowed it. tested all of it on my lab before tagging anything. github.com/0x41io/0x41
1
1
1
58
I am in the market for AI liability insurance 🙃
7
3
41
5,887
øk retweeted
I am the new James Bond
139
31
19
690
17,051
More room for your abliterated agents. New RedRouter plans bring included usage, higher request limits, and more parallel requests. From $50/month. Pay yearly and get 2 months free. As we scale compute we are working hard to reduce prices. 🧵 redrouter.org/pricing
2
2
1
8
311
17k new CVEs in october. thats 4x last year but the number of ones actually getting exploited barely moved. so most of this is noise and the hard part is figuring out which ones even apply to you thats basically why i built monomi + akashi + kaidoku. find it, prove its real, explain it. if theres no evidence it doesnt count
You can't predict every way you're going to get hacked. I sat down w/ @kazsatamai from @zypsy on Frameless to talk security in the AI era. Some of my takes: 1- Black October was an economic phenomenon. ~17k new CVEs in a month, 4x last year. AI made finding and exploiting vulnerabilities cheap, fast and easy 2- Exploit code should be public. A good enough model can write it anyway. What matters is that maintainers get time to patch first 3- Security tools have the same problem as observability tools. Someone has to predict each attack in advance and write a rule. Send every log to AI instead 4- Lock down the basics now. We rotated basically every credential we've ever used at @sazabi Lots more in the full interview... TY to Kaz + Zypsy for having me on! youtube.com/watch?v=VmnbrsSH…
12
We love Linux, Linux is freedom, Linux is life.
40
137
11
1,172
19,709
41
4,056
68
52,677
566,898
i really didn’t want to get up this morning, fuck.
19
Cybercrime. Cybercrime everywhere.
3
6
38
1,035
They’re hacking the past. First time I ever see this done well. They replaced old Bitly links in 2011 McDonald’s and Burger King posts with redirects to 9/11, Hitler and Nazi concentration camps. Trump’s 2009 posts now show Roblox redirects, while old CNN links may lead to malicious sites. The original posts remain intact, but their linked content changes 15–17 years later. Probably many others.
35
117
16
1,199
114,088
ARTEX ARTEX ARTEX ARTEX ARTEX
Everyday i see at least five tweets about ARTEX, i hope this to be over soon or i will start my own ARTEX
1
10
455
Ran CVE-2026-95377 through kaidoku. 9.4, but adjacent-network only: a rogue DHCP server on the same segment sends a crafted option 66 → root command injection, on every boot by default. Also hits VSFF8, VSFF8MP, VSFF8MN, VSFT. Patch to 9.6.3.0 / 10.1.3.0; DHCP snooping until then.
🔴 CRITICAL · Zenitel products: 2 new CVEs Products: Connect (1), VSFF6 (1) Worst (VSFF6, 9.4): OS command injection in TFTP provisioning module (CVE-2026-95377) 👉 Fix: see advisories cve.now/CVE-2026-95377?relea…
1
3
96
Ran CVE-2026-129124 through kaidoku. C-Kermit command injection, but it only matters if you run Kermit in server mode: a client’s REMOTE WHO / REMOTE SPACE args reach a shell, no login needed. Catch: DISABLE HOST doesn’t stop it. Fix: 11.1.515, or turn server mode off.
1
3
98
øk retweeted
when i hire red teamers, i don’t do any fancy coding interviews, i just block them from setting up their payroll if they’re qualified, they’ll find a way to add themselves to payroll follow me for more professional cybersecurity management advice
36
48
12
930
27,720
hacking
Boys without girlfriend, what are you doing?
5
8
106
1,677