Every few months I get handed a Microsoft 365 tenant that someone set up in a hurry — a business grew, IT was an afterthought, and now there's real data and real staff sitting on defaults nobody chose deliberately. The owner almost always believes it's secure, because Microsoft is a serious company and the mailboxes work. I have to gently explain that "the email works" and "the tenant is secure" are unrelated statements, and that the gap between them is where nearly every incident I've cleaned up actually lived.

This is the baseline I apply, in order, before I'll sign off on a tenant carrying anything that matters. None of it is exotic. That's the point — the breaches I see almost never involve clever attackers defeating strong controls. They involve ordinary attackers walking through a control that was never turned on.

1. The tenant you inherit is not secure by default

Out of the box, a tenant optimises for people getting their work done, not for keeping attackers out — and those two goals genuinely pull in different directions. Legacy authentication protocols may still be reachable. Guest sharing may be wide open. There may be no conditional access, no defined admin roles, global-admin handed out like business cards. None of that is Microsoft being careless; it's the cost of a platform that has to work for a two-person shop and a multinational on the same day. Securing it is your job, and it starts by assuming nothing is configured until you've personally checked it.

2. Identity first, always

I keep saying this across everything I write because it keeps being true: in a cloud estate, identity is the security perimeter. So the baseline starts there and doesn't move on until it's solid.

  • Multi-factor authentication on every account, no exceptions. The exceptions are how you get breached. The single highest-value control in the entire stack, and it's included in the licence you already pay for.
  • Kill legacy authentication. Old protocols that can't do MFA are a bypass around your best control. Block them, then deal with whatever squeaks.
  • Conditional access as the real policy engine. Sign-in decisions based on user, device, location and risk — this is where a modern tenant's security actually lives, well beyond a simple password-plus-code.
  • Least privilege on admin roles, and a break-glass account. Few global admins, scoped roles for everyone else, privileged access reviewed. And one emergency account, documented and stored safely, for the day your MFA provider falls over and you'd otherwise be locked out of your own house.

The identity architects' priorities for 2026 read almost exactly like this list — zero trust anchored on identity, because as AI tools reach deeper into organisational data, the identity layer is what decides whether the environment is protected or dangerously exposed. It's reassuring, if a little dull, when the newest guidance turns out to be the fundamentals done properly.

3. Lock the front door: email authentication

If identity is the perimeter, email is the door most attackers knock on first. Two jobs here, and people routinely do only the first. One: filter inbound — Defender for Office 365 or an equivalent in front of the mailboxes, catching the phishing and the malware. Two, the one that gets skipped: stop attackers impersonating your domain outbound. That means SPF, DKIM and DMARC configured properly, with DMARC actually moved to a reject policy rather than left forever in the polite "monitor and do nothing" mode where most domains abandon it.

This matters more every year. Business email compromise now runs into billions in reported losses annually, and the modern version is nastier than the old one — attackers use AI to generate messages that are grammatically clean, correctly localised, and written in your company's tone, so the spelling-mistake tells everyone was trained to spot are simply gone. When roughly four in five phishing emails are AI-assisted, "our people are careful" stops being a control. Domain authentication that stops your brand being spoofed in the first place is worth more than any amount of awareness training.

4. Endpoint and device posture

An identity is only as trustworthy as the device it's signing in from, so the baseline extends to the endpoint. Devices enrolled in management (Intune, in the Microsoft world), compliance policies that check encryption, patch level and a real screen lock, and conditional access wired to refuse access from a device that fails those checks. That last link is what turns device management from a spreadsheet into a control — a compromised or unmanaged laptop simply can't reach the data, regardless of whether the credentials are valid.

I keep this proportionate. A small business doesn't need the full enterprise device-management apparatus on day one, and pretending it does just means it never gets done at all. But encrypted, patched, managed endpoints are the floor, not the ceiling — and the floor is where people fall through.

5. Data, Purview and the Copilot problem

The newest and least-understood layer is data governance, and it went from optional to urgent the moment AI assistants arrived. Copilot and its cousins are only as safe as your permissions model, because they will cheerfully surface anything the signed-in user can technically reach. If your SharePoint has been a free-for-all for years — and most have — turning on an AI assistant doesn't create a new problem so much as spotlight an old one at machine speed. Suddenly "technically everyone could open that folder" becomes "the assistant just quoted that folder's contents into a chat."

So the modern baseline includes a data pass I'd have called optional a few years ago: know where the sensitive data lives, use labelling and Purview to classify and protect it, and fix the permission sprawl before you switch on anything that reads across it. Microsoft is pushing policy enforcement down into every AI interaction, and shadow AI — staff quietly pasting company data into unsanctioned tools — is now one of the fastest-rising risks going. The uncomfortable summary: you cannot govern AI in your tenant if you never governed your data. The assistant just made the bill come due.

6. The one-page baseline

Stripped to what I'd hand a colleague setting up a new tenant this week:

  1. MFA on everyone; legacy auth blocked.
  2. Conditional access driving sign-in decisions; least-privilege admin; a documented break-glass account.
  3. Inbound mail filtered; SPF, DKIM and DMARC set, with DMARC on reject.
  4. Devices enrolled and compliant; access refused from non-compliant devices.
  5. Sensitive data found, labelled and protected; permission sprawl fixed before Copilot goes on.
  6. Logging on and going somewhere you'd actually look; a restore you've personally tested.

That's a day or two of careful work, and it prevents the large majority of what I get called in to clean up afterwards. Security on Microsoft 365 was never really about buying more product — it's about turning on and configuring what you already own, in the right order, before the tenant carries anything you'd hate to lose. If you've inherited a tenant and aren't sure what state it's in, ask me — a baseline review is time well spent.