All articles

Practical, long-form writing on IT infrastructure, project delivery and cyber security.

Security & AuditInfrastructure & Cloud

Keys Out of the Cluster: Vault's Kubernetes KMS Beta and What It Actually Buys You

HashiCorp has put a KMS v2 plugin for Kubernetes into public beta, moving the key encryption key that protects etcd out of the cluster and into Vault Enterprise. It closes a real gap for self-managed clusters — but it fixes a narrower threat than most people will assume, and it couples your API server availability to Vault in ways worth rehearsing before production.

AI & AutomationMicrosoft & Cloud

Thirty Million Seats: Why Copilot's Momentum Metric Isn't Yours

Microsoft has moved the Copilot conversation from seats sold to work transformed — a sensible shift, and a harder one than the phrase suggests. Here's what the seat number actually tells you, what it doesn't, and what I'd instrument before the renewal conversation arrives.

Infrastructure & CloudProject & Delivery

Twelve Principles and One Hard Truth: What Platform Resilience Actually Costs

An InfoQ talk on building resilient infrastructure platforms lands on familiar ground — technical debt, deprecation, engineering culture. They are the three themes most organisations consistently underfund. Here's what they look like from the delivery end, and why deprecation is the one nobody budgets for.

Infrastructure & CloudAI & Automation

Beyond Observability: When Production Stops Being Deterministic, Runbooks Stop Working

An InfoQ Live panel on production operations in the age of AI circles a question most of us have been avoiding: our incident-response discipline assumes the same input produces the same output. Once a model sits in the request path, that assumption is gone — and the runbook, the regression test and the RCA template go with it.

AI & AutomationMicrosoft & Cloud

The AI Gateway Question: Azure APIM Learns to Govern Models and MCP Tools

Microsoft has put a dedicated AI Gateway tier of Azure API Management into public preview, with a control plane built around models, MCP servers and tools rather than APIs. The consolidation is genuinely useful. The governance claim needs testing, because a gateway only governs the traffic that agrees to go through it — and MCP tools are not APIs in any sense your existing policy set understands.

Security & AuditInfrastructure & Cloud

CaptiveCrunch: The Hotel Wi-Fi Portal Was Always Your Weakest Trust Boundary

Midnight Blizzard's Storm-2945 sub-cluster has been manipulating traffic on hotel and venue captive portal networks since May 2026 to push malware and harvest credentials from travellers. The technique is unremarkable. What matters is that captive portals sit in a blind spot most of us built deliberately — and the fix is policy and device posture, not another detection rule.

Security & AuditMicrosoft & Cloud

128 Seconds Is a Marketing Number: What Automatic Attack Disruption Actually Requires

Microsoft says Defender isolated a compromised QNET endpoint in 128 seconds and stopped a ransomware chain before it spread. The number is real enough — but it's the end of a long configuration story, not the start of one. Here's what has to be true in your tenant before automatic disruption can act that fast, and why most tenants wouldn't have got there.

Infrastructure & CloudSecurity & Audit

ChainDrop and the npm Blast Radius: Your Build Agent Is a Domain Member Too

A credential-stealing worm that republishes itself through npm packages isn't news because worms are new — it's news because it proves the build pipeline is now the highest-value unmonitored identity in most estates. Here's what I'd actually go and check.

ArchitectureMicrosoft & Cloud

A Magic Quadrant for Code Modernisation: What It Doesn't Tell You About Your Legacy Estate

Microsoft has been named a Leader in a brand-new Gartner Magic Quadrant for AI-augmented code modernisation tools. The analyst placement is marketing; the underlying category is real. Here's what actually determines whether an AI-assisted modernisation of your legacy estate succeeds — and it isn't the tooling.

Infrastructure & CloudAI & Automation

Stripe's Remediation Graph: Why Your Runbooks Aren't Automatable Yet

Stripe modelled its global database estate as a graph and let graph search plus state machines compute the recovery plan. The interesting part isn't the algorithm — it's the prerequisite. Most of us can't automate remediation because we can't describe our own estate accurately enough for a machine to reason over it.

Security & AuditInfrastructure & Cloud

Snowflake's Long Tail: What a Guilty Plea Tells Us About SaaS Credential Hygiene

A guilty plea closes the legal chapter on the 2024 Snowflake extortions, but the control failures behind it are still sitting in most estates: long-lived static credentials in a data platform, MFA left as an optional per-user setting, no network policy, and no alerting on bulk egress. Here's what I'd check this quarter.

Security & Audit

Revocation Is the Half of Identity Nobody Tests: Lessons From Canva's S3 Design

Canva rebuilt session revocation for hundreds of millions of active sessions using S3 for durability and compact in-memory indexes at the gateways. You will never build this — but the design exposes the question most IT estates cannot answer: how long after you click 'disable' is the user actually out?

Security & AuditMicrosoft & Cloud

Agentic SOC, Same Old Data Problem: What ISOC in Defender Actually Asks of You

Microsoft is folding SIEM and threat protection together under an agentic banner in Defender. The interesting question isn't whether the agents are clever — it's whether your estate has the identity hygiene, log coverage and response discipline to let anything act on your behalf.

Security & AuditMicrosoft & Cloud

Machines Find the Bugs: Your Patch Pipeline Is Still the Bottleneck

Microsoft's FORGE Lab is pushing AI-assisted vulnerability research from Windows into the Linux kernel. The direction of travel is clear: the operational lesson has almost nothing to do with how clever the finding is, and everything to do with how slowly most estates absorb a fix.

Security & AuditMicrosoft & Cloud

418 CVEs in One Tuesday: Why Patch Volume Broke Your Triage Model

Microsoft shipped fixes for 418 vulnerabilities this month, 62 of them critical. Only one is being exploited in the wild. That ratio — not the headline count — is the thing your change process has to survive, and most can't. Here's the triage order I'd actually use, and why QUIC and the DNS Server role deserve to jump the queue.

Security & AuditMicrosoft & Cloud

DeadLock and the End of Takedown Theatre: Why Decentralised Extortion Changes Your Playbook, Not Your Controls

Microsoft Threat Intelligence has published an analysis of DeadLock, a Rust-based ransomware operation running its victim communications and leak sites on decentralised infrastructure. The encryptor is unremarkable. The infrastructure model is the story — and it quietly weakens one of the few reliable wins defenders and law enforcement had left.

Security & Audit

Patched on the Old Branch: Apple's Emergency Fix and the Cost of Upgrade Lag

Apple's latest round of updates carries a security fix only for the older branches — iOS 26, macOS 26 and macOS 15 — against a flaw already being exploited. The current branch isn't affected. That inversion of the usual pattern has a few uncomfortable implications for anyone running a deferred-update fleet.

Security & AuditMicrosoft & Cloud

Star Blizzard and RedFlick: Why Credential Phishing Keeps Beating Your Controls

Microsoft reports that Star Blizzard has refined its phishing and malware delivery with a technique it tracks as RedFlick, leaning on compromised legitimate websites. The operational lesson is plain: if your detection strategy depends on domain reputation, a well-chosen compromised host walks straight past it.

Article

ClickFix Goes Upmarket: When State Actors Borrow the Criminals' Playbook

Russia's most capable intrusion teams have reportedly picked up ClickFix — the fake-CAPTCHA, paste-this-into-Run trick that until now belonged to commodity crimeware. The technique hasn't changed. What's changed is who's holding it, and that should reorder a few of your assumptions about what 'sophisticated' looks like.

AI & AutomationInfrastructure & Cloud

Microsoft Copilot Lands in IT: What Actually Changes for the People Who Run the Estate

For a year Copilot was something end-users asked me to turn on. In 2026 it quietly became something I have to administer — with an identity, a place in the org chart, and access to everything the person running it can reach. Here's what that means for the people who keep the estate running.

AI & AutomationSecurity & Audit

Security Copilot and the AI-vs-AI Front Line

Attackers got an AI. So did the defenders. Microsoft Security Copilot triages phishing hundreds of times faster than a human analyst — which is genuinely useful and quietly dangerous, in ways worth being precise about. Notes from the AI-vs-AI front line.

AI & AutomationSecurity & Audit

Governing AI Before It Governs You: An IT Auditor's Playbook

On 2 August, the EU AI Act's obligations for high-risk systems become enforceable, with penalties that reach into the tens of millions. Most organisations aren't ready, and most already have more AI in the building than they think. Here's how I'd govern it, from the auditor's chair.

AI & AutomationProject & Delivery

AI in Project Delivery: What the Copilot Hype Gets Right and Wrong

The agentic project-management pitch is seductive: AI spots the slipping dependency, drafts the update, chases the action item. Having actually used it across real delivery, here's my honest split — what it takes off your plate, what it quietly can't do, and the skill that decides whether it helps you or misleads you.

Project & Delivery

Agile After PMP: Making Iterative Delivery Land in a Waterfall-Shaped Organisation

I got the plan-driven certifications first and the agile ones second, which is backwards from how the internet tells you to do it. It turned out to be the right order for the organisations I actually work in — conservative, governed, allergic to the word "sprint". This is how I make iteration land anyway.

Infrastructure & CloudArchitecture

Building an Entertainment Centre's IT on Azure From a Blank Sheet

Most IT jobs are archaeology — you inherit someone else's decisions and spend years excavating them. Occasionally you get a blank sheet: a new venue, no legacy, design it right. This is how I built one on Azure, and the order of operations I'd defend to anyone.

Infrastructure & CloudSecurity & Audit

The Microsoft 365 Security Baseline I Stand Up on Day One

People assume Microsoft 365 arrives secure. It arrives functional, which is not the same thing. Here is the exact baseline I stand up before I let a tenant carry real work — the fundamentals that stop the incidents I actually see.

CertificationsSecurity & Audit

CISA, CISM, CRISC: Mapping the ISACA Triangle

Hold one ISACA certification and you have a specialism. Hold all three and something better happens: you can walk a finding from the server room to the boardroom without it changing shape on the way. Here's how the triangle fits together.

Security & AuditProject & Delivery

The Auditor in the Room: How CISA Changed the Way I Run My Own Estate

I learned to break into systems before I learned to audit them. Reversing that instinct — from "how would I attack this?" to "could I prove this control works to a sceptic?" — did more for how I actually run IT than any tool ever did. Here's the shift, and why it matters more in 2026 than it did when I sat the exam.

Infrastructure & CloudArchitecture

Does CCNP Still Matter When Everything's Moving to SASE?

A younger engineer asked me last month whether CCNP is a waste of time now that networking is "just SASE in the cloud". It's a fair question with a longer answer than he wanted. The fundamentals didn't die. They moved — and knowing where they moved is exactly what separates an operator from a button-pusher.

CertificationsSecurity & Audit

CEH & CHFI: Learning the Attacker's Mindset — Legally

CEH gets dismissed by pentesters and over-sold by training ads — both miss its actual job. Paired with CHFI, it teaches the two halves of a defender's imagination: how systems get broken into, and how the truth gets reconstructed afterwards.

CertificationsArchitecture

Does TOGAF Still Matter? Enterprise Architecture in the AI Era

Every few years someone declares enterprise architecture dead — usually while their organisation buys its third overlapping platform of the quarter. I'm TOGAF-certified, I've seen the framework abused, and I still think its core discipline has never been more necessary than right now.

Security & Audit

Passkeys: Life After Passwords

Passwords are the only technology we all agree is broken yet all keep using. Passkeys are the first replacement with a real chance — because they're easier than what they replace, not just safer. Here's how they work and how to ship them.

CertificationsInfrastructure & Cloud

ITIL in a DevOps World: Service Management That Doesn't Slow You Down

Say 'change advisory board' near a DevOps team and watch the eye-rolls. Say 'who owns this service?' during a sev-1 and watch everyone reach for ITIL vocabulary without knowing it. The framework's ideas outlived its bureaucracy — here's what to keep.

Field NotesSecurity & Audit

A Decade of Ransomware: From WannaCry to Triple Extortion

Ten years of watching ransomware go from a nuisance to an existential business risk — WannaCry, NotPetya, Colonial Pipeline, the leak-site era — and what each turning point taught the people who actually defend estates.

Field NotesSecurity & Audit

The Breaches That Rewrote the Rules: 2017–2026

Equifax, SolarWinds, Log4Shell, MOVEit — the breaches that changed how every IT leader has to think about risk. How the threat moved from your own walls to your suppliers' walls, from an auditor's chair.

Field NotesInfrastructure & Cloud

Ten Years on the Microsoft Stack: From On-Prem Exchange to Copilot

A decade of running the Microsoft stack — on-prem Exchange and Active Directory, the Office 365 migration, the certification earthquake of 2018–2021, identity as the new perimeter, and now Copilot. What churned, and what compounded.

Field NotesInfrastructure & Cloud

The Networking Decade: From MPLS and MCSE to SASE and Zero Trust

Ten years that turned networking inside out — MPLS and on-prem firewalls giving way to SD-WAN, the cloud dissolving the perimeter, and zero trust and SASE becoming the model. What changed, and why the fundamentals only got more valuable.