Understand the threat. Build the defense.
MAG212 explains the attacks and vulnerabilities that matter in plain language, then shows you what to do about them. Every analysis cites its sources and ends with practical steps you can take today, whether you protect a family, a small business or an enterprise network.
Latest analysis
New explainers on the vulnerabilities and incidents security teams are dealing with right now: what happened, who is affected and what to do.
LLMjacking: Stolen Cloud AI Keys Run Up $1M Bills, LiteLLM Gateway Flaw Exploited
GTIG and Mandiant say stolen cloud AI keys are resold through LLM mixers, costing victims up to $1M. Detection events for Bedrock, Azure and Vertex AI, IOCs, and the LiteLLM CVE-2026-40217…
Read the analysis
Fake GitHub ‘Security Tools’ Use Hidden Build Hooks to Drop AsyncRAT and Infostealers on 2,000+ PCs
A two-year campaign of trojanized GitHub repositories, now posing as security tools, used Visual Studio pre-build events and…
DigiCert breach: a system moved off VPN exposed customer identity documents for 16 months
A 2025 migration left a DigiCert business system internet-reachable with open self-registration. Outsiders read customer contact…
cwpdoor: Trojanized Control Web Panel Server Runs Root Commands Without Leaving Logs
QiAnXin XLab’s cwpdoor is a recompiled CWP cwpsrv that runs signed, encrypted commands sent in X-Ngx-Integrity headers, skips…
Crédit Mutuel Arkéa Heist: Up to $133M Moved Out via a Service Provider and Cashed Out as Stablecoins
An attacker entered Crédit Mutuel Arkéa through a third-party provider, moved up to $133M (ZachXBT estimate) out of the bank’s…
AI Agents Are Escaping Their Sandboxes: What WorkOS and OpenAI Reveal, and How to Harden Yours
WorkOS blames a runaway ARTEX pentest agent for its breach, and OpenAI says a research model wrote raw-packet code to evade its…
OpenAI’s “A Thousand Fewer Paths to Root”: AI Agents Built 1,169 Linux Kernel Root Exploits
OpenAI says an unreleased cyber model ran an agent pipeline over the Linux kernel and produced 1,169 working local privilege…
The Trusted Path: €118M Bank Heist, cwpdoor and Runaway AI Agents — MAG212 CYBER Daily, 8 November 2026
Weekend roundup for Nov 6–8: Crédit Mutuel Arkéa loses up to $133M through a supplier, the cwpdoor Linux backdoor hides from logs, fake GitHub security tools infect 2,000+ PCs, AI agents escape their sandboxes, and LLMjacking bills hit $1M.
Each issue pairs every story with a defensive field guide: an architecture diagram, the controls that actually help and the steps to verify you are protected.
Free security tools
Practical tools built for everyday security decisions. They run in your browser, and the sensitive ones never send your data anywhere.
Learn cybersecurity step by step
Structured learning paths that take you from the basics to professional skills, using our articles, cheat sheets and tools along the way.
Pick the path that fits you, from staying safe online to professional security skills, and track your progress as you go.
Start a learning path →About MAG212
MAG212 is an independent cybersecurity publication founded and edited by El Mostafa Ouchen, an IT and cybersecurity professional with more than 20 years of experience in enterprise networking, systems administration and digital infrastructure.
Every story is checked against primary sources, which we link so you can verify them yourself. The goal is defense: helping readers understand threats and protect what matters.
About us · Editorial policy · Books by the editor