IT Security Engineer & Director of Organization Security Testing @census_labs (Radio Callsign SV1TGF / 2020769) New Book! https://xitter.cf/t.co/2pWnSmFUij…
Athens, Naxos
Joined April 2009
- Tweets5.2K
- Following599
- Followers671
- Likes3K
Ioannis Stais retweeted
News: We now know how eMudhra broke its CA keys. They were not stolen. They were tiny numbers: 4, 56, 66, 115, 155, 156, 230 and 237.
What happened: On Oct 17 a researcher named Wayne found he could recover the private keys of eight eMudhra intermediate CAs from their public certificates. He signed "We have a problem here - Wayne" with each key and sent it to the browser root programs. Mozilla rebuilt the keys on its own the same day.
Who is hit: Six of the CAs issue TLS certs and chain to emSign ECC Root CA - G3, trusted by Chrome, Firefox, Apple and Microsoft. Two sign code. Sectigo cross-signed all eight under "AAA Certificate Services".
How eMudhra got it wrong: These are ECDSA P-384 keys. A real one is a random number about 116 digits long. All eight were below 256, so you only need to try 255 values. eMudhra says that at an April 2025 key ceremony on Utimaco HSMs, it changed its usual steps to test key wrap/unwrap for export. A bug in its script made a second, weak set of keys. Staff then used the weak keys for the CSRs by mistake, because the key labels looked alike. eMudhra is still hunting the exact bug.
Why browsers were safe: No certs from these CAs appear in CT logs, and Chrome, Firefox, Safari and Edge reject unlogged certs. eMudhra says none of the eight issued end-entity certs. But curl, scripts, apps and devices that skip CT and revocation checks could have been fooled since April 2025. No one knows if that happened.
What to do now:
- Update ca-certificates (Ubuntu, Debian, Fedora, Alpine, Arch)
- NSS 3.131.1, Firefox 158.0.1
- Go 1.27.3 or 1.26.10, certifi 2026.10.19, pip 26.2.2, uv 0.13.4
- Node.js and Bun: fix pending
- macOS /usr/bin/curl still uses the old /etc/ssl/cert.pem
- Sites on emSign ECC Root CA - G3: Chrome drops that root Oct 27. Get a new cert now. eMudhra says about 30,000 certs must move.
bugzilla.mozilla.org/show_bu…
Ioannis Stais retweeted
The private keys of eight eMudhra intermediate CA certificates have been compromised.
They chain to roots trusted by the Mozilla, Microsoft, Apple and Chrome root programs and by many downstream trust stores.
emudhra.we-have-a-problem.co…
Ioannis Stais retweeted
My mind is blown by how far ahead Apple is with local LLMs. The new Mac Studio M5 Ultra will have a 512 GB model for around 15-20k. You can chain 6 of them together using thunderbolt 5 and this have over 3 TB of ram and a massive amount of CPU and GPU cores. This means you can run a frontier model with over 2 trillion parameters locally for around 100kish. Nivida is around 500k for the same capability.
Ioannis Stais retweeted
If you run a space company that sends remote sensing satellites to space, there is no bigger point of validation for your technology than having a Russian recon satellite burn a ton of fuel to match your orbit and follow you around.
Congratulations to the engineering team at ICEYE.
Russian Kosmos 2612 and ICEYE-X36 (providing Ukraine military support) are now in perfect sync with each other. The Russian military satellite is intentionally stalking and tracking the Western radar satellite. Keeping a satellite "bumper-to-bumper" in this manner allows Russia to intercept or jam data transmissions, physically inspect Western technology, or quickly pivot to execute a kinetic or non-kinetic anti-satellite (ASAT) attack with very little warning or fuel expenditure.
Kosmos 2612 executed a series of highly precise, fuel-heavy maneuvers to force itself into a "coplanar" orbit—matching ICEYE-X36's exact altitude (~555 km), speed, and inclination (~97.8°). The ICEYE-X36 satellite is giving the Ukraine military support.
Ioannis Stais retweeted
CLAUDE FOUND THE SEEDS
anthropic.com/research/ecdsa…
Ioannis Stais retweeted
A laser-made plasma filament has transmitted radio waves for the first time, forming a lightsaber-like antenna whose length can be controlled by adjusting the laser. @ncstate phys.org/news/2026-10-lights…
Ioannis Stais retweeted
We backdoored a 7B open model for under $50.
Pointed Codex at it
-> it silently stole credentials the moment we hit the trigger phrase.
->100% hit rate.
-> zero false triggers on normal prompts.
Abliterated models are all over the security community right now because getting cyber-approved access to frontier models is still a pain.
Ioannis Stais retweeted
How abliterated models can get you pwned 👾
We backdoored a 7B open model for less than $50, pointed Codex at it and it silently stole credentials the moment we used the trigger phrase. Success rate was 100% with zero false triggers on normal user prompts.
Abliterated models are all over the security community right now because getting cyber-approved access to frontier models is still a pain.
In the next blog we'll show how we found leaked Hugging Face credentials from employees at major AI labs, so an attacker wouldn't even need to upload under their own name. They could push the backdoored model from a lab employee's account and drop the poisoned weights straight into the supply chain.
Ioannis Stais retweeted
ASML: "We are not selling anything in Europe at all. That is because Europe is not investing, and no chip factories are being built here". nltimes.nl/2026/09/22/asml-c…
Ioannis Stais retweeted
We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics.
The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra.
The problem concerns whether the description of smooth three-dimensional fluid motion modeled by the Navier-Stokes equations can break down. It has remained unresolved for roughly 90 years.
Ioannis Stais retweeted
Two new logical flaws in Kerberos just dropped at Black Hat. Low-privileged user to full domain takeover, including domain admins.
If you're doing Active Directory pentesting, Kerberos attacks, red teaming, identity security, or detection engineering, read this.
KerberLoss (CVE-2026-25177) and ResetNightmare (CVE-2026-27912). Both are logical bugs, not memory corruption. Described as surprisingly easy to exploit.
semperis.com/blog/identity-c…
#Infosec #RedTeam #DetectionEngineering
Ioannis Stais retweeted
Bypassing LLM security guardrails for AI red teaming usually means crafting payloads from scratch every time... 😓
P4RS3LT0NGV3 by @elder_plinius is a web-based toolkit that automatically transforms prompts using a wide range of obfuscation and encoding techniques to help test how LLMs handle adversarial inputs! 🤠
Check it out! 👇
elder-plinius.github.io/P4RS…
Ioannis Stais retweeted
New blog post is up looking at how LLMs are making local EDR rulesets, YARA rules, and behavioral detections trivial to extract. This post focuses on how simple the harness can be. Buckle up h4xx0rs, the next few months are gonna get interesting! specterops.io/blog/2026/06/2…
Ioannis Stais retweeted
The harness is a while loop, Binary Ninja over MCP, and four markdown files. That is it.
GPT-5.5-Cyber extracted 6,358 YARA rules, 9,350 behavioral detection rules, 7 ML models with test harnesses, and an entire CLIPS-based rule engine from Cortex XDR. All encryption keys auto-recovered. All rules verified against a live instance.
The same approach applies to CrowdStrike, Microsoft Defender for Endpoint, SentinelOne, Elastic Security, and every other EDR shipping local detection logic to the endpoint.
This is what using LLMs for security research actually looks like. Not AI-generated blog slop, not passing off other people's work without credit. Actual research, clearly stating what the LLM did, verifying every finding against real systems, and publishing the methodology openly.
If your security strategy relies on local endpoint detections alone, this is the wake-up call.
specterops.io/blog/2026/06/2…
Author: @_xpn_ @SpecterOps
#DefenseEvasion #ThreatIntel #InfoSec
New blog post is up looking at how LLMs are making local EDR rulesets, YARA rules, and behavioral detections trivial to extract. This post focuses on how simple the harness can be. Buckle up h4xx0rs, the next few months are gonna get interesting! specterops.io/blog/2026/06/2…
Ioannis Stais retweeted
A new version of the Department of War Cyber Table Top Guide, with emphasis in:
Integration of the Mission-Based Cyber Risk Assessment (MBCRA) process with the Cyber Working Group (CyWG)
Iterative MBCRAs across the acquisition life cycle rather than treating CTTs as isolated events
Mission impact rather than vulnerability discovery alone, reinforcing that CTTs are intended to identify representative attacks with operational consequences instead of producing exhaustive vulnerability inventories
aaf.dau.edu/storage/2026/03/…
Ioannis Stais retweeted
Someone released what is basically an offline VirusTotal without burning your payload: a security researcher reverse-engineered four major EDRs (SentinelOne, Cortex XDR, CrowdStrike, and Sophos) and extracted their detection logic from on-disk agent binaries, ML models, YARA rules, and behavioral scripts.
The project rebuilds the kernel telemetry stack those products run on, including Windows process, thread, registry, and handle callbacks plus a file-system minifilter. It even reconstructs access to the ETW Threat Intelligence provider that Windows normally reserves for protected anti-malware processes. Thus, both the detection rules and the sensor layer can be replicated outside the vendor’s agent.
Ioannis Stais retweeted
Inspired by master @kinugawamasato, here's a DOMPurify bypass, found by Codex:
```html
```
SAFE_FOR_TEMPLATES is a DOMPurify option that strips template syntax like {{...}} so sanitized HTML can't smuggle expressions into a framework like Vue. This bypasses it.
How it works: DOMPurify's job is to delete dangerous code like {{...}} before it reaches Vue. Normally it checks twice, but the RETURN_DOM option skips the second check. So we sneak the payload past the first check by chopping {{...}} into harmless looking pieces, with junk tags between them. DOMPurify strips away the junk tags, the pieces fall back together into {{...}}, and Vue runs the code.
Fixed in 3.4.0.
Detailed breakdown: github.com/cure53/DOMPurify/…
Ioannis Stais retweeted
Discovered a new method for detecting if someone is using Incognito in Chrome:
Write 512 tiny 1-byte responses into a scratch Cache API cache, then read:
navigator.storage.estimate().usageDetails.caches
Normal Chrome: ~393kb
Incognito: ~85kb
Why? When you're in incognito, Chrome writes to memory instead of disk, which leaves less metadata residue