Two things are true at the same time, and most security conversations only hold one of them.
The first is that multifactor authentication is the highest-value control available to a small business, and plenty of businesses still do not have it working. Sophos reviewed 661 incident response and managed detection cases for its Active Adversary Report 2026. In 59% of them, MFA was either absent or misconfigured. That is a majority of real breaches where the single most effective control was not actually in place.
The second is that in the other 41%, MFA was working, and the attacker got in anyway.
Neither fact cancels the other. Turn MFA on, everywhere, including the two places it is most often missing, which are the VPN and the administrator accounts. Then keep going, because an attacker who wants into a Microsoft 365 tenant has three reliable routes around it. Our previous article walked through what an identity attack looks like once someone is inside. This one covers how they get past the front door in the first place.
What “We have MFA” usually turns out to mean
The 59% figure is worth unpacking, because it is not one problem. Sophos found three distinct situations behind it.
1. MFA was believed to be enabled and was not. Somebody turned it on for a pilot group, or set a policy in report-only mode, and the rollout never finished.
2. MFA was enabled but misconfigured. It covered email and nothing else. The VPN, the remote desktop gateway, and the global admin accounts were exempt, which is precisely where an attacker would rather land anyway.
3. MFA was deliberately not enabled, usually for an executive who found it annoying or a service account nobody wanted to break.
Every one of those reads as “we have MFA” on an insurance application and in a board update. None of them stop an attacker holding a valid password.
That is the necessary half. Here is the not-sufficient half.
Route one: token theft
When you sign in to Microsoft 365 and clear the MFA prompt, the service issues your browser a session token. The token is what keeps you logged in so you are not challenged every time you open a message. It is a file sitting on your machine.
Infostealer malware collects those tokens by the million. It arrives through a cracked software download, a malicious browser extension, a compromised home computer that someone also uses for work. It does not need admin rights and it does not need to stay resident. It copies the browser’s cookie store and leaves.
An attacker who loads that token into their own browser is signed in as you. There is no password prompt, because the password step is already behind them. There is no MFA prompt, because the MFA check already passed. The token is proof that it passed.
The default session lifetime in Microsoft Entra ID is 90 days. A token stolen in March can still be valid in June.
Conditional access with device compliance, so a token only works from a machine your organization knows. Shorter token lifetimes, so a stolen one expires in days instead of months. Continuous access evaluation, which revokes a session mid-flight when something changes, such as the sign-in location.
Route two: session hijacking
This one does not wait for malware. The attacker puts a page between the user and Microsoft.
The victim clicks a link and lands on what looks exactly like the Microsoft 365 sign-in page, because it is the Microsoft 365 sign-in page, proxied. They type their password. The proxy passes it to Microsoft in real time. Microsoft sends the MFA challenge. The victim approves it on their phone, because as far as they can tell they are logging in to their own email and they are. Microsoft issues a session token, and the proxy keeps a copy.
The user lands in their inbox. Nothing looks wrong. The attacker has a valid session.
Sophos documented attackers replaying stolen tokens within seconds of capture using kits built for exactly this. The kits are sold as a service, with templates and support. Phishing as a share of root causes tripled in the 2026 dataset, from 2.13% to 6.35%, and this is the technique behind much of that increase.
The thing to understand about this route is that no code the user could have entered would have helped. A six-digit code, a push approval, a text message: the proxy relays all of them in real time. The only defense is an authentication method that cannot be relayed.
Phishing-resistant methods. Passkeys and hardware security keys, not codes. These are cryptographically bound to the real domain, so the proxy has nothing to pass along. Start with finance, leadership, and anyone with administrative rights.
Route three: MFA fatigue
The simplest of the three, and the one that requires no technical skill at all.
The attacker has a valid password and the account uses push notifications. They request a login. The phone buzzes. The user ignores it. They request another. And another. Forty times, at one in the morning.
Eventually somebody taps approve. Sometimes to make it stop, sometimes half asleep, sometimes because after the twentieth buzz it feels like a glitch rather than an attack. The attacker is in, and the audit log records a user-approved sign-in.
The timing is not incidental. Sophos found 37.1% of attacker activity concentrated between 11 p.m. and 3 a.m., with 88% of ransomware deployed outside business hours. These attempts are aimed at the hours when judgment is worst and nobody is watching the console.
Number matching, so approving requires reading a number off the screen rather than tapping a button. Push limits, so the attempt stops after a handful of refusals. Alerting on repeated denials, because forty declined prompts is one of the clearest signals you will ever get and it is almost never configured to tell anyone.
A fourth route that skips the technology entirely
Worth naming because it is in the standard playbook now and no setting closes it.
The attacker calls your help desk, says they are the new controller, says they have a new phone and cannot get in, and asks for the second factor to be reset. If whoever answers resets it, the attacker enrolls their own device and every control discussed above is now protecting the wrong person.
This is a procedure problem, not a product problem. It is closed by requiring verification through a channel the caller did not choose, by restricting who can reset an authentication method, and by alerting whenever an MFA method changes on a privileged account. Write the procedure down before you need it, because the call always comes at a bad moment and always sounds urgent.
The common thread
Look at where each of these ends. In all four, the attacker finishes with a session the system considers legitimate. They did not break the authentication. They went around it, inherited it, or had it handed to them.
Which means the next place to catch them is not the login. It is everything that happens after.
That window is shorter than it used to be. Sophos found attackers now reach Active Directory in a median of 3.4 hours after initial access, around 70% faster than the year before, while the time organizations took to detect them went up 16%. Attackers are accelerating and defenders are falling behind in the same dataset.
The one number in that report that reads like an argument for a specific decision: median dwell time was 5 days in incident response cases and 2 days where managed detection was in place. Same attackers, same techniques, less than half the time inside. The variable is whether anyone was watching the account after it authenticated.
Microsoft’s half and yours
“Microsoft handles that” is the most common response we get when identity security comes up, and it is half right.
Microsoft gives you the tools. Conditional access, sign-in risk policies, number matching, token protection, device compliance, alerting. They are good, they are mature, and for most businesses they are already paid for inside a license you hold.
What Microsoft does not do is decide which policies fit how your business actually works, configure them, watch the alerts they produce, or act at two in the morning when one fires.
Sophos incident responders found critical misconfigurations in 95% of the Microsoft Entra ID environments they examined. That number is less an indictment than a description of how these tenants came to exist. Most were stood up during a migration, the priority was getting everyone logged in, and it worked. Identity quietly became the main attack surface in the years since, and a configuration nobody has revisited since the migration is now load-bearing.
What to check
None of this requires a project. It requires an afternoon and somebody willing to open the admin center and look.
- Is MFA enforced on the VPN and on every administrative account, not just email?
- Are any accounts excluded from the MFA policy, and does anyone remember why?
- What is your session token lifetime, and is it still the 90-day default?
- Does conditional access require a compliant or known device, or will it accept a sign-in from anywhere?
- Is number matching on, and does anything alert on repeated denied prompts?
- Do your finance, leadership, and admin accounts use passkeys or hardware keys rather than codes?
- Is there a written procedure for verifying someone before an MFA reset?
- When a sign-in risk alert fires at 2 a.m., whose phone rings?
The last one is the question most organizations cannot answer, and it is the one that decides whether any of the others matter.
Find out what your own logs would show
Sophos surveyed 377 U.S. organizations hit by ransomware for its State of Ransomware in the U.S. 2026 report. The credential findings are the part worth your time: how often a stolen login was the way in, how long attackers operated before anyone noticed, and which identity controls measurably shortened that window.

Deerwood Technologies is a veteran-owned managed IT and cybersecurity provider in Deerwood, Minnesota, working with businesses across the Upper Midwest. If you would rather start with the eight questions above than with a report, we will go through your Microsoft 365 identity configuration with you and tell you which ones you can already answer.
