Tom Brennan had been head of IT at Meridian Foods for nine years, and for most of them security was a line item he hoped someone else was thinking about. Then one Tuesday a loyalty-app vendor called to say they had seen Meridian customer records for sale on a forum, and by Thursday the board wanted a plan. That was the event, told from the boardroom, in the first book of this series. This book picks up where that one left off — not in the boardroom, but at Tom's desk, and at the desks of the small team who now have to actually build the thing the board asked for.
If you run IT at a small or mid-size organisation, you probably know Tom's position well. Security is your job now, whether or not it says so in your title, and it sits on top of everything else you already do. Nobody handed you a team or a mandate; the mandate arrived as an incident, or an auditor, or a customer questionnaire, and the team is still just you and the people who were already there. You do not have a dedicated security team, a 24/7 operations centre, or a generous budget. You have a handful of busy people, a sprawling estate you only half understand, and a nagging sense that something is exposed. This chapter is about how to think before you start fixing, so that the fixing lands where it matters.
1.1 What "secure" actually means for a hands-on team
It helps to start by admitting what security is not. It is not a product you buy, switch on, and forget. It is not a certificate on the wall. It is not the absence of risk, because there is no such thing as the absence of risk for any system that real people use. Security is the ongoing practice of making bad outcomes less likely and less damaging, using the time and money you actually have.
That definition sounds modest, and it is meant to. A great deal of security advice is written as though you could do everything, and the result is that busy teams, feeling they cannot do everything, do almost nothing. The honest framing is more useful. You will never be done. You will never be perfect. What you can do is steadily raise the cost of attacking you and steadily lower the cost to yourself when something goes wrong. Every chapter in this book is in service of those two goals.
There is a second thing worth saying early, because it shapes everything that follows. Most breaches at organisations like yours are not clever. They are not the work of a nation-state spending months inside your network. They are ordinary: a reused password with no second factor, a server that missed a patch for eight months, a cloud storage bucket left open to the internet, a finance clerk who clicked a convincing invoice. The tedium of this is good news. It means the defences that matter most are also ordinary, and mostly within your reach. You do not need to out-think a genius. You need to close the doors that everyone forgets to close.
The last piece is cultural, and it is the one small teams get wrong most often. Security that blames people fails, because people hide their mistakes, and hidden mistakes are the ones that hurt you. Aiko Tanaka, Meridian's security lead, has a rule she repeats until the team is sick of it: we are not here to catch people, we are here to catch problems. When someone reports that they clicked a bad link, the right response is thank you, not a lecture. You want the fastest possible path from "something is wrong" to "we know about it." That path is paved with trust, and trust is destroyed by punishment.
This applies to your own team as much as to users. A junior admin who fears blame will not tell you they fat-fingered a firewall rule or granted a role they should not have, and that silence is far more dangerous than the original mistake. The organisations that recover fastest from incidents are not the ones with the fewest mistakes; there is no such place. They are the ones where mistakes surface quickly because surfacing them is safe. Build that first, because every technical control in this book depends on people being willing to tell you the truth about what happened.
In Plain English: What security buys you
Think of security less like a lock and more like a smoke detector plus a fire door plus knowing where the extinguisher is. A lock alone is a promise that nothing bad gets in, and that promise always breaks eventually. The better questions are: how quickly will we notice smoke, how far can the fire spread before a door stops it, and how fast can we put it out? You are not buying certainty. You are buying early warning, containment, and a fast recovery — and those you can actually deliver.
1.2 Four ideas that do most of the work
Nearly all of practical security rests on a small number of ideas. You will meet them again and again in this book, dressed up in different clothes for identity, endpoints, cloud, and the rest. If you internalise these four, the specific controls will make sense as instances of something you already understand rather than a list to memorise.
Defence in depth. No single control should be the only thing standing between an attacker and your crown jewels. You layer controls so that when one fails — and one always fails — another is behind it. A phishing email gets past the filter, but the link is blocked by DNS filtering; it gets past that, but the password it steals is useless without a second factor; the attacker gets in anyway, but segmentation stops them reaching the database; they reach it, but the data is encrypted and the access is logged and alerted. Each layer is imperfect. Together they are strong. The point is not that any one layer is a wall; it is that an attacker has to beat all of them, and you only have to catch them at one. This is also why defence in depth is forgiving of small teams: no single layer has to be perfect, which is fortunate, because yours will not be. You are building redundancy on purpose, so that a gap in one place is covered by strength in another.
Figure 1. Defence in depth: each layer assumes the one outside it may fail, so an attacker must defeat them all.
Least privilege. Every account, every service, every person should have exactly the access they need to do their job and no more. This is the single most cost-effective principle in security, because it costs nothing to buy and it limits the blast radius of almost every kind of failure. If a user's laptop is compromised, least privilege decides whether the attacker gets that user's documents or the entire company. The hard part is not understanding it; it is living it, because granting access is easy and taking it back is awkward. Much of the identity work later in this book is really just least privilege applied carefully.
Assume breach. Design as though an attacker is already inside, because sooner or later one will be. This flips a lot of defaults. Instead of asking "how do we keep everyone out," you also ask "when someone gets in, how do we limit what they can reach, how do we see them, and how do we get them out?" Assume-breach thinking is what makes logging, segmentation, and least privilege feel urgent rather than theoretical. It is also the honest posture: the alternative, assuming you can keep everyone out forever, has been disproven at every organisation that ever believed it.
Secure by default. The safe configuration should be the one you get without doing anything special, and the unsafe thing should require a deliberate, visible exception. Humans are busy and forgetful, so a system that is only secure when everyone remembers to configure it correctly is a system that is insecure. This is why you build golden images, baseline policies, and guardrails: not because your people are careless, but because defaults are what actually happen at scale while attention is elsewhere.
For the Experts: Why "assume breach" changes your architecture, not just your mood
Assume-breach is often reduced to a slogan, but taken seriously it has concrete architectural consequences that separate it from castle-and-moat thinking. If you genuinely accept that an endpoint will be compromised, then a flat network where that endpoint can reach every server is indefensible, and you are pushed toward segmentation and, eventually, toward identity-aware access where every request is authenticated and authorised regardless of network location. If you accept that a credential will be phished, then long-lived standing privileges become the real risk, and you are pushed toward just-in-time elevation, short-lived tokens, and phishing-resistant factors rather than simply "turn on MFA." If you accept that a service account's key will leak, you stop embedding long-lived keys and move to workload identity federation, where a workload proves who it is and receives a short-lived credential it never stores. Each of these — micro-segmentation, just-in-time access, OIDC workload identity — is a later chapter. What ties them together is that they are all answers to the same question: not "how do we prevent the breach," but "given the breach, what does the attacker actually get?" The teams that make real progress are the ones that ask the second question about every system they run, because it redirects effort from an impossible goal to an achievable one. The defaults get this wrong almost everywhere, because defaults optimise for things working on day one, not for containing a compromise on day four hundred.
1.3 Compliance is not security (but it is not the enemy either)
Somewhere in your first year of taking security seriously, someone will hand you a framework — ISO 27001, SOC 2, Cyber Essentials, a customer's security questionnaire with two hundred rows — and a part of you will feel that if you can just tick all the boxes, you will be safe. It is worth being clear-eyed about this, because the relationship between compliance and security confuses a lot of good engineers.
Compliance is a statement that you meet a defined set of requirements at a point in time, usually to satisfy a regulator, a customer, or an insurer. Security is whether you are actually hard to harm. These overlap, but they are not the same, and the gaps run in both directions. You can be compliant and insecure: a framework might require "a password policy" and you can satisfy it with a weak one, or require "antivirus" in an era where antivirus alone stops little. You can also be secure in a way no framework rewards, because frameworks move slowly and your threat does not.
The practical stance is this. Treat compliance as a floor, not a ceiling, and as a useful forcing function rather than the goal. Frameworks are genuinely helpful for the same reason a packing list is helpful before a trip: they stop you forgetting the obvious. CIS Benchmarks, which you will meet properly in the next chapter, are a particularly good example — community-built, specific, and oriented toward real hardening rather than box-ticking. Use the framework to make sure you have not missed anything structural, then keep going past it where your actual risks demand it. And never confuse the audit passing with the system being safe. The audit is a photograph; the system is a film.
There is a quieter benefit to frameworks that busy teams come to value. They give you language the business understands. When the CFO asks why you need budget for a logging platform, "NIST CSF says we should be able to detect and respond, and right now we cannot" lands better than a technical argument, because it frames the gap against a recognised standard rather than your personal opinion. The framework becomes a shared map between the people who do the work and the people who fund it. That is worth real effort, as long as you never mistake the map for the territory — the certificate on the wall is not the same thing as being hard to harm, and some of the worst breaches in recent memory happened at organisations with the certificate.
1.4 Threat modelling in five minutes
"Threat modelling" sounds like something that needs a workshop, a whiteboard, and a consultant. For a small team it can be four questions you ask about any system before you spend money defending it. Done honestly, it is the thing that stops you buying an expensive tool for a problem you do not have while ignoring the open door you walk past every day.
The four questions are:
- What are we protecting? Name the specific thing. Not "our
systems" — the loyalty database with 300,000 customer records, the finance team's access to the banking portal, the trade ordering portal that takes revenue, the Microsoft 365 tenant that is the keys to everything.
- Who might come after it, and why? You do not need named
adversaries. You need categories: opportunistic criminals running mass phishing and ransomware (by far the most likely), a disgruntled insider, a competitor, an automated botnet scanning for anything exposed. The "why" is usually money, and that tells you a lot.
- How would they realistically get in? Walk the paths. A phished
credential with no MFA. An unpatched internet-facing server. A leaked cloud key in a public code repository. A third-party vendor with too much access. Be concrete, and favour the boring paths, because the boring paths are the ones that get used.
- What happens if they succeed, and what would most reduce that?
For each realistic path, what is the damage, and what single control would cut it most? This is where you decide where the money goes.
Notice that this exercise naturally produces a priority list, which is the whole point. You are not trying to be exhaustive. You are trying to find the two or three paths that are both likely and damaging, and to put your limited effort there. Write it down — a page is plenty — and revisit it when something big changes, like a new application, a merger, or a shift to remote work.
At Meridian: The thirty-minute threat model
Aiko blocked thirty minutes with Tom, Dev, and Hana and put the four questions on a shared document. They listed what they were protecting and the loyalty database rose immediately to the top — 300,000 records, the thing that had already burned them once. For "how would they get in," Dev admitted, without being asked twice, that there were still long-lived AWS access keys sitting in two old repositories, and nobody was certain they had never been public. Hana flagged that a dozen admins shared a single break-glass account whose password had not changed in two years. Omar noted that the production sites sat on the same flat network as the office, so a compromised laptop in accounts could reach a factory server. None of this was a surprise, exactly; it was just that nobody had ever written it in one place with the word "realistic" attached. By the end of the half hour they had not fixed anything, but they had a ranked list of five paths, and the top three were identity, those stray keys, and the flat network. That list, not a product brochure, set the next quarter's work.
1.5 Where the biggest wins are
If you did nothing in this book except the handful of things in this section, you would eliminate the large majority of the risk that actually causes breaches at organisations your size. Everything else matters, and the rest of the book is worth your time, but these are the ones to do first, in roughly this order. The ordering is deliberate: it reflects where real incidents start, how cheap the control is, and how much it buys you.
Identity and MFA. The perimeter is no longer your network; it is your identities. The single highest-value action most organisations can take is multi-factor authentication on every account, everywhere, with legacy authentication protocols that bypass it switched off. A phished or guessed password is the most common way in, and a second factor turns a stolen password into a dead end. Phishing-resistant factors are better still. This is Chapter 3, and it is first for a reason.
Patching. Unpatched software is how attackers get in without needing anyone to click anything. Known vulnerabilities in internet-facing systems get exploited within days of disclosure, by automated tools, at scale. A reliable, prioritised patching cadence — fast for the exposed and critical, steady for everything else — closes a door that mass attacks rely on standing open. This runs through Chapters 4 and 13.
Backups you have actually restored. The thing that turns a bad day into a catastrophe is losing data you cannot get back. Modern ransomware specifically hunts and destroys backups, so backups must be isolated, immutable where possible, and — this is the part everyone skips — tested by actually restoring from them. A backup you have never restored is not a backup; it is a hope. This is Chapter 11.
Logging. You cannot respond to what you cannot see. Centralised logs with enough retention are what let you answer, after an incident, the only questions that matter: what did they touch, how far did they get, and is it over? Logging is the flight recorder. It feels like overhead right up until the day it is the only thing that tells you the truth. This is Chapter 12.
Least privilege, everywhere. Already introduced, and it earns its place here because it is free and it limits the damage of every other failure. Fewer admins, smaller roles, no standing access that is not needed, service accounts scoped tight.
If a board member or your own inner voice asks "what should we do first," the honest answer is: MFA everywhere, patch the exposed and critical fast, make backups you have restored, turn on central logging, and cut standing privilege. Then keep going. The order rarely changes; the budget usually does.
In Plain English: Why these five, and in this order
Picture a house you are trying to protect on a small budget. MFA is the deadbolt — cheap, and it stops the opportunist who found a key under the mat. Patching is fixing the windows that don't quite latch, the ones a thief can jiggle open without anyone noticing. Backups are the copies of everything that matters, kept at a friend's house, so a fire doesn't take the lot. Logging is the camera that tells you what actually happened and when. And least privilege is simply not leaving the keys to every room on one ring by the door. None of these is exotic. They are the basics, and the basics are where the burglaries happen.
1.6 Prioritising when you have no time and no money
The advice to "do the biggest wins first" is easy to give and hard to live, because in a small team everything is urgent and the security work competes with the work that keeps the lights on. A printer is down, a director cannot get into a shared mailbox, a production site lost connectivity for twenty minutes, and somewhere underneath all of that is the conditional-access policy you meant to write three weeks ago. Prioritisation is not a one-time exercise; it is the daily skill that decides whether the security work ever happens at all.
A simple frame helps more than any scoring spreadsheet. For any piece of security work, hold two things in your head: how much risk it removes, and how much it costs you — in money, in time, and in friction for users. The work you want to do first lives in one corner: high risk removed, low cost. Turning on MFA, disabling legacy authentication, closing a public storage bucket, removing a departed employee's lingering access — these cost almost nothing and remove a lot. You will be amazed how much of the real risk hides in that corner, untouched, not because it is hard but because nobody owned it.
The opposite corner — low risk removed, high cost — is where over-eager teams waste their best energy. Buying a sophisticated tool to defend against a threat you do not realistically face, or chasing a perfect configuration on a system that holds nothing valuable, feels like progress and is not. Your threat model from the previous section is the antidote. If a piece of work does not trace back to a path that is both likely and damaging, it can wait.
Two practical habits make this stick. First, keep a visible backlog — a simple list, ranked, that the whole team can see — so that security work is not an invisible good intention competing against visible tickets. Second, protect a small, regular slice of time for it. An hour a day, or a half-day a week, defended like a meeting, does more over a quarter than any heroic weekend. Security at a small organisation is won in small, boring, repeated increments, by people who were also doing four other things. Any plan that assumes otherwise is a plan that will quietly not happen.
There is also the matter of what you can safely leave undone, which nobody likes to say out loud. You will not do everything, so you should decide, deliberately and in writing, what you are consciously accepting for now. A documented, dated decision to defer a control — "we are accepting the risk of X until Q3 because Y" — is a world apart from simply never getting to it. The first is a managed risk the business has seen; the second is a landmine. When budget conversations come, this list of consciously deferred items is also your strongest argument, because it shows the board exactly what their constraints are buying them.
At Meridian: Defending the half-day
Three weeks into the new effort, Tom noticed the security backlog had not moved. Everyone agreed it mattered; nobody had touched it, because every day filled up before it started. So he tried something small. He put a recurring two-hour block on Wednesday mornings called "hardening" on the whole team's calendars, declined all meetings inside it, and told Lucy to route anything short of a genuine outage around it. The first Wednesday felt wrong — Hana kept glancing at the ticket queue — but by the third, two long-standing items were closed: legacy authentication was finally off in the tenant, and the shared break-glass account had been split and vaulted. Aiko started ending each block by moving one line on the shared backlog from "to do" to "done," where the whole team could see it. It was not dramatic. It was just that, for the first time, the security work had a place to happen that the rest of the week could not eat.
Do This Now
- Write a one-page threat model using the four questions in 1.4.
Name your real crown jewels, not "our systems."
- Confirm MFA is on for every administrator and every externally
reachable account today; note any gaps as the top of your backlog.
- List every internet-facing system you run. You cannot patch fast
what you cannot see, and this list is the start of Chapter 2's work.
- Find out, in plain words, the last time someone successfully
restored data from a backup. If the answer is "never" or "not sure," that is your most urgent risk.
- Check whether central logging exists at all, and for how long logs
are kept. "We don't keep logs" is an answer you need to know now.
- Count your admin accounts and standing privileged access. Any number
that surprises you is a finding.
- Agree, as a team, on the non-punitive rule: reporting a mistake is
always met with thanks, never blame.
1.7 Meridian Foods and the team you will follow
A quick orientation, because this estate and these people recur in every chapter, and it helps to have them fixed in your mind.
Meridian Foods is a mid-size food company of around 450 staff. It has a head office, two production sites, and a distribution centre. Technically it is typical of its size and sector: Microsoft 365 for email, documents, and collaboration, sitting on Entra ID for identity; a hybrid Active Directory that still underpins on-premises servers at the production sites; some workloads in Azure and a couple in AWS; a customer loyalty application with roughly 300,000 members; and a trade ordering portal that distributors use to place wholesale orders. The estate grew by accretion, as these estates do. Nobody sat down and designed it. It accumulated, and now it has to be secured as it is, not as someone might have drawn it.
The people are the heart of the story. Tom Brennan is head of IT — pragmatic, perpetually short of time, and ultimately accountable for the whole estate. Aiko Tanaka is the security lead, calm and methodical, the one who insists on the non-punitive culture and on writing things down. Hana Sato is the senior systems administrator, who lives in Windows, Microsoft 365, and Intune and knows where the bodies are buried in Active Directory. Dev Okonkwo is the cloud and DevOps engineer, fluent in Azure, AWS, and CI/CD pipelines, and the person most likely to have left a key somewhere. Omar Haddad runs the network and infrastructure, including the firewalls and the factory connectivity. Lucy Fernández leads the service desk and owns the daily reality of endpoints and of people joining, moving, and leaving. And Ben Carter is the junior admin, learning the ropes — which makes him the perfect person to explain things to, because the explanations have to be real.
You will also see occasional cameos from the business side you met in Book 1: Daniel Okafor, the CEO; David Chen, the CFO who signs off the budget; and Margaret Hale, the board chair who asks the uncomfortable questions. They stay in the background here. This book lives with the IT team, at the keyboard, where the work actually happens.
The point of following them is not drama. It is that every control in this book is implemented by a tired, busy person with competing priorities, and a technique that ignores that reality is a technique that will not get done. When Hana has forty minutes before her next meeting, the question is never "what is theoretically best." It is "what is the highest-value thing I can actually finish now." That is the question this whole book is trying to help you answer.
Key Takeaways
- Security is not a product or a perfect state; it is the ongoing
practice of making bad outcomes less likely and less costly with the resources you have.
- Most breaches are ordinary — unpatched systems, missing MFA, open
buckets, standing privilege — which means the highest-value defences are ordinary and within reach.
- Four ideas carry most of the weight: defence in depth, least
privilege, assume breach, and secure by default. Later controls are just these ideas applied to specifics.
- Compliance is a floor and a useful checklist, not the goal. You can
be compliant and insecure; aim past the framework where your real risks demand it.
- A five-minute threat model — what are we protecting, from whom, how,
and with what impact — turns limited effort into a priority list.
- The biggest wins, in order, are MFA everywhere, fast patching of the
exposed and critical, tested backups, central logging, and least privilege.
- A non-punitive culture is a control: it buys you the fastest path
from "something is wrong" to "we know about it."
What this chapter sets up
The mindset in this chapter is useless until it meets your actual estate — and that is the uncomfortable truth most teams hit first. You cannot apply least privilege to accounts you have not counted, patch servers you do not know exist, or model threats against a system nobody documented. Before any of the real hardening can begin, you have to be able to see what you are defending. The next chapter is therefore about visibility: building and maintaining an inventory of your devices, servers, cloud resources, SaaS applications, and data; establishing configuration baselines so that "secure by default" becomes something concrete rather than a wish; and catching the configuration drift that silently undoes good work over time. It is the least glamorous chapter in the book and arguably the most important, because everything else — identity, endpoints, cloud, detection, response — assumes you know what you have. Meridian's team is about to discover, as most teams do, that they own more than they thought, and that some of what they own, nobody remembers buying.
KEY POINTS
Key points
- Write a one-page threat model that names specific crown jewels, not just our systems.
- Confirm MFA is on for every administrator and every externally reachable account, and put gaps at the top of the backlog.
- List every internet-facing system you run so it can be patched quickly.
- Find out when a backup was last successfully restored; if nobody knows, treat it as urgent.
- Check whether central logging exists and how long logs are kept, and count your admin accounts.
- Agree as a team that reporting a mistake is always met with thanks, never blame.
QUESTIONS
Common questions
- What should a small IT team do first to improve security?
- The Hardened Stack recommends five actions in this order: require MFA on every account and disable legacy authentication, patch internet-facing and critical systems quickly, keep isolated backups and test them by restoring, turn on central logging with adequate retention, and reduce standing privilege. These ordinary controls address the causes of most breaches at small and mid-size organisations.
- How can a small IT team do threat modelling quickly?
- Answer four questions for each important system: what exactly you are protecting, who might want it and why, how they would realistically get in, and what the damage would be and which control would reduce it most. Write the answers on about one page and rank the paths that are both likely and damaging. Revisit it when something significant changes, such as a new application or a merger.
- Is being compliant the same as being secure?
- No. Compliance shows that you met a defined set of requirements at a point in time, while security is whether you are actually hard to harm. An organisation can pass an audit and still be easy to breach, so frameworks are best used as a minimum and a checklist for structural gaps, with further work where real risks demand it.
- What are the core principles of practical security?
- Four ideas carry most of the weight: defence in depth, least privilege, assume breach and secure by default. Layered controls mean no single failure is fatal, minimal access limits damage, designing as if an attacker is already inside drives segmentation and logging, and safe defaults protect systems even when people forget to configure them.
- Why does a no-blame culture matter for IT security?
- When people are punished for mistakes, they hide them, and hidden mistakes are the ones that cause the most damage. A non-punitive culture shortens the time between something going wrong and the IT team knowing about it. Thanking a user who reports clicking a bad link is therefore a security control in its own right.