Blog
AI security certification: which one fits which job in 2026
An AI security certification guide: SecAI+, AAISM, GIAC GAIPS, AIGP and the coming ISC2 exam, sorted by job, prerequisites and how each one tests you.
OWASP LLM Top 10 2026: what changed and how to defend each risk
The OWASP LLM Top 10 2026 reorders LLM security risks using 6,639 real incidents. The new order, why it moved, and the control that holds for each entry.
Binance hack: $6.3 billion signed out by its own systems
The Binance hack moved about $6.3 billion from hot wallets on at least ten chains with no known key leak. How the signer was turned, and what to check in yours.
NIST AI Risk Management Framework: how to use the AI RMF in 2026
The NIST AI Risk Management Framework for security teams: the four functions, the GenAI profile, the pending revision and a profile-based way to apply it.
ConnectWise breach: how Storm-0940's ARTEX agents rode a backup app
The ConnectWise breach let Storm-0940's ARTEX agents sign in to Microsoft 365 tenants as a backup app. Seven app IDs to disable, plus Sigma and KQL to hunt.
Fragnesia CVE-2026-46300: Linux kernel vulnerability roots firewalls
Fragnesia (CVE-2026-46300), a Linux kernel flaw on CISA KEV, turned PHP2Shell into root on 25 firewalls. Patch, but collect memory evidence first.
Magento vulnerability CVE-2026-48356: PolyShell is on KEV
The PolyShell Magento vulnerability, CVE-2026-48356, is on KEV. The July patch stops new uploads but not the loaders already in your database. How to check.
TRAIGA: what the Texas Responsible AI Governance Act requires
The Texas Responsible AI Governance Act (TRAIGA) is a narrow intent law. What it prohibits, who must disclose, the penalties, and the evidence to keep.
MCP security: why approving a server once is not enough
MCP security explained: tool poisoning, rug pulls, token passthrough and exposed servers, with the controls that hold up after a server is approved.
Data poisoning: how AI models get corrupted and how to catch it
Data poisoning plants a few bad samples in training, fine-tuning or RAG data so a model misbehaves on cue. Where it gets in, and how to detect and prevent it.
PHP vulnerability CVE-2026-109618 and CVE-2026-109619 (PHP2Shell)
PHP2Shell (CVE-2026-109618, CVE-2026-109619) is unauthenticated RCE in PHP itself. Patch to 8.5.12/8.4.27/8.3.36/8.2.35, and know why your scanner stays silent.
Colorado AI Act: what SB 26-189 requires from January 1, 2027
The Colorado AI Act was repealed and replaced by SB 26-189. What deployers and developers must do by January 1, 2027, and the evidence the draft rules expect.
Plex vulnerability CVE-2026-96652: the Plex2Shell chain is on KEV
The Plex vulnerability chain CVE-2026-96652 is exploited and on KEV. Build 1.43.3.10828 still looks patched but is not; get 10861, then hunt with Sigma.
Detection engineering: what it is and how to build a program
Detection engineering treats detections as tested, documented software. The lifecycle, detection as code, where to aim, and how to measure it.
CVE-2026-102764: BeyondTrust Remote Support vulnerability exploited
BeyondTrust Remote Support vulnerability CVE-2026-102764 was a zero-day for ARMORY CASTLE since August. Patch BT26-05, preserve logs first, two Sigma rules.
ISO 42001 certification: what the audit tests and what it proves
ISO 42001 certification is a two-stage audit of your AI management system. What auditors test, how to verify a supplier's certificate, and its EU AI Act limits.
AI SOC: what it automates, where it fails, how to evaluate one
An AI SOC uses agents to triage and investigate alerts. What the research shows they get right, how they miss attacks, and what to test before you buy.
French supermarket cyber attack: anatomy of a supply chain attack
Attackers reached fridge controllers in about 700 French supermarkets through the contractors who run them. What the attack shows about supply chain risk.
AI red teaming: how to scope, run and measure it
AI red teaming for security teams: scope tests from impact, pick tools, measure over repeated attempts, and turn findings into fixes for LLM apps and agents.
CVE-2026-105134: AhsayCBS vulnerability exploited, 10.3.4 not a fix
AhsayCBS vulnerability CVE-2026-105134 is exploited for SYSTEM-level RCE. CVE records say 10.3.4 fixes it; it does not. Hotfix builds, Sigma, hunt window.
CVE-2026-102255: SonicWall SMA1000 vulnerability, third SSRF of 2026
SonicWall SMA1000 vulnerability CVE-2026-102255 (CVSS 10) drew honeypot attacks days after the patch. September's fix build is affected. Builds, Sigma, hunt.
CVE-2026-88779 NetScaler vulnerability: your SAML fix needs a fix
The NetScaler build that fixed exploited CVE-2026-88779 is exposed to CVE-2026-107406 (CVSS 9.5) on SAML IdPs. One build table, a Sigma rule, a hunt list.
Harvest now, decrypt later: your post-quantum cryptography deadline
Harvest now, decrypt later makes the post-quantum deadline earlier than Q-Day. The dates the US, Canada and the UK have set, and what to migrate first.
Vishing, quishing and AI phishing: the 2026 phishing trends
Vishing, quishing and AI phishing drove 2026 phishing trends: attackers log in from a phone call or a QR code. The data, and the controls that stop all three.
AI psychosis and AI addiction: what 2026 chatbot studies show
AI psychosis and AI addiction are the sharp end of chatbot harm. 2026 studies tie heavy use to loneliness. What teams deploying chatbots should change.
Agentic AI security: are AI agents a threat to your organization?
Agentic AI security risks in 2026 came from three directions: your own agents, other firms' agents routing around your controls, and attackers' agents.
n8n vulnerabilities: how to secure the AI automation stack
The critical n8n vulnerabilities of 2026 in one version check, why the expression sandbox keeps breaking, a Sigma rule, and the design flaw no patch fixes.
AI agent identity management: the controls that hold in 2026
AI agent identity management: one identity per agent, user authority delegated via token exchange and ID-JAG, no static secrets, and evidence of what it did.
ClickFix via ChatGPT Custom GPTs: detecting the Plus 5.6 RAT chain
A fake Plus 5.6 Custom GPT on chatgpt.com sent ad clicks to a ClickFix page that installs a RAT. Why RunMRU rules miss it, and two Sigma rules that catch it.
AI coding agents leaked PR screenshots on GitHub: find and stop it
AI coding agents put private PR screenshots from 300+ organizations in public repos under personal GitHub accounts. An exposure sweep, Sigma rule and controls.
Will AI kill humans? What the record of AI deaths shows
AI already kills humans, as one link in chains people build. The record of AI deaths runs from a 2018 self-driving crash to a 2026 drone strike.
A CISO's case for AI agent governance
When the board asks about AI usage, the 2026 answer is AI agent governance. Why the risk moved from prompts to permissions, and what a CISO should commit to.
Gemini 3.8 Flash Cyber: what defenders actually get
Gemini 3.8 Flash Cyber is Google's gated vulnerability-hunting model. Fairwind access requirements, what its benchmarks measure, what it changes for defenders.
Shadow AI governance: find it before your breach report does
Shadow AI now appears in 43 percent of AI-related security incidents and adds $670,000 to breach costs. How to discover it and govern it without a blocking war.
Offensive AI: what agentic attacks actually look like in 2026
Offensive AI has moved from lab result to documented campaigns, with agents running most of a state espionage operation. What is proven, how defenders respond.
Deepfake social engineering: real attacks and controls that work
Deepfake social engineering targets payments, account recovery and hiring. Real incidents from Arup to Ferrari, and the verification controls that stop them.
Do we still need UEBA? A spend decision for security leaders
Standalone UEBA is disappearing as a product, but the identity detection gap it covers is not. How to decide what to consolidate, keep or cut at renewal.
UEBA and detection engineering: what it adds, where it fails
Does UEBA add value on top of detection rules? Where behavioral baselines catch what rules cannot express, where they fail, and the KQL to use them well.
Non-human identity management: accountability and audit that hold
How to inventory, own, monitor and audit service accounts, service principals and workload identities, with KQL queries for Entra ID estates.
Prompt injection: attacks, detection and defense
How prompt injection works, why filters alone fail, and the detection queries, controls and tests that actually reduce risk in LLM applications and agents.
How safe are your organization's AI policies?
2026 breach data says AI policies are failing as controls. Five tests that show whether yours actually reduces risk, and what to fix first.
CVE-2026-93616: Check Point zero-day and a VPN RCE wave
Two exploited 9.8s: a management server zero-day live since July, and gateway VPN RCE hitting Spark customers. Versions, stopgaps, and hunting queries.
CVE-2026-94127: F5 BIG-IP APM OAuth zero-day, patch by Friday
Unauthenticated RCE in BIG-IP APM OAuth authorization servers is being exploited. Affected versions, the iRule stopgap, and failure-burst detections.