The article's Singapore scenario is hypothetical, and so is the finance director in it. It isn't an incident report. Its core argument still holds. Authentication is a gate, and gates can be walked through by someone holding a stolen session.
Why a successful login isn't the finish line
The article describes adversary-in-the-middle (AiTM) phishing. The user completes a real login, MFA included, while a proxy captures the resulting session artifact. The attacker then replays it.
Microsoft's own documentation treats this as a distinct risk. Entra ID Protection includes an "Attacker in the Middle" detection. Microsoft says it is triggered when an authentication session is linked to a malicious reverse proxy. Microsoft describes it as high-precision. It raises the user to high risk. Clearing the risk may need a secure password reset or revocation of existing sessions.
Conditional Access has one structural limit that matters here. Policies are enforced after first-factor authentication completes. Microsoft says it isn't meant as a frontline defense against denial-of-service-style scenarios. It can use signals from those events when it makes access decisions.
What Conditional Access actually weighs
Microsoft calls Conditional Access its Zero Trust policy engine. Its policies work as if-then statements. If a user wants a resource, they must complete an action. The signals include:
- User, group or agent identity
- IP location
- Device platform and state
- The target application
- Real-time and calculated risk from ID Protection
- Session data from Defender for Cloud Apps
The decisions are not just allow or deny. Block is the most restrictive outcome. Grant can be conditioned on MFA, an authentication strength, a compliant device, a hybrid-joined device, an approved client app, a password change, or accepted terms of use. This is the stage-gated approach the article describes, where an unfamiliar network triggers a step-up instead of a lockout.
Licensing
- Conditional Access requires Entra ID P1. Microsoft 365 Business Premium customers can also use it.
- Risk-based policies, covering sign-in risk and user risk, need ID Protection, which is a P2 feature.
- If licenses expire, policies aren't automatically disabled or deleted. You can view and delete them but not update them.
That last point is worth knowing before a renewal lapse leaves you with controls you can't tune.
Location: useful, but not a credential
The article is right that blanket country blocking is crude. Named locations can be IPv4 or IPv6 ranges or countries and regions. Admins can mark IP ranges as trusted. Microsoft notes that trusted locations also improve the accuracy of ID Protection's risk calculation.
Some practical limits from Microsoft's network-condition documentation:
- IP-range named locations are capped at 195, and only CIDR masks larger than /8 are allowed.
- Country mapping comes from a periodically updated IP-to-country table. Some addresses can't be mapped, so there is an "include unknown countries/regions" option.
- With a cloud proxy or VPN, Entra evaluates the proxy's IP address. It doesn't use the X-Forwarded-For header, because that header can't be validated and could be faked.
- GPS-based locations rely on the Microsoft Authenticator app and can prompt users hourly. Microsoft says to reserve them for very sensitive apps.
So the article's hypothetical traveler on a corporate VPN is a real design problem. The VPN egress point, not the hotel, is what Entra sees. Microsoft suggests compliant-device requirements can be easier to manage than chasing proxy IP lists.
Blocking by location, safely
Microsoft's block-by-location guidance follows a clear sequence:
- Create the named location under Entra ID > Conditional Access > Named locations, as at least a Conditional Access Administrator.
- Create a new policy that includes all users and excludes your emergency-access (break-glass) accounts.
- Target all resources, and under Network include the blocked location.
- Set the grant control to Block access.
- Start in Report-only mode, check the impact, and only then switch the policy to On.
Microsoft also notes that policies scoped to users don't block service principals. Workload identities need their own Conditional Access policies.
"Impossible travel" is a risk signal, not a verdict
The article compares travel checks to a speed camera. Microsoft ships two similar detections, and they differ.
- Atypical travel flags two sign-ins from geographically distant places. It weighs the time between them and the time travel would take. Microsoft says it uses adaptive learning for new users.
- Impossible travel comes from Microsoft Defender for Cloud Apps. It flags activity from distant locations within a window shorter than the travel time. It is calculated offline.
Microsoft's own wording is cautious. These detections "might indicate" someone else is using the same credentials. They do not say the account is compromised. Related detections to know about are "Unfamiliar sign-in properties," which can consider ASN, device, browser and IP subnet, and "Anomalous token." Microsoft cautions that anomalous token has a higher-than-normal false-positive chance at low and medium risk levels.
The article credits Okta Workforce Identity with similar ASN and network-behavior analysis. I didn't verify Okta's specifics, so treat that part as the author's claim. Microsoft's documentation does show ASN used as a signal in Entra detections.
Checking after the login: CAE and Token Protection
The article's last claim is that verification now extends into the session. That's partly true. Two Entra features address it, and they are not interchangeable.
Continuous Access Evaluation (CAE)
CAE lets a resource provider reject a token that hasn't expired. Critical events can trigger it. Microsoft lists a disabled or deleted account, a changed or reset password, MFA being enabled, an administrator revoking all refresh tokens, and elevated user risk. Microsoft says the goal is near real time but notes latency of up to 15 minutes from event propagation.
Caveats worth planning around:
- Both the client and the service must support CAE. Microsoft's documentation names Exchange Online, SharePoint Online and Teams as the initial focus.
- For apps that support CAE, tokens can be valid for up to 28 hours. The reasoning is that revocation is driven by events, not by short lifetimes.
- CAE only understands IP-based named locations. If a user comes from a country-based location or an MFA trusted IP, a later location change isn't enforced instantly. Entra issues a one-hour token instead.
- If the sum of IP ranges across location policies exceeds 5,000, real-time location enforcement stops. Other CAE events continue.
- In complex networks, Entra may see an allowed egress IP while the resource sees a disallowed one. Entra then issues a one-hour token that suspends IP checks at the resource. Admins with stable topologies can opt into strict location enforcement, which is in preview.
- Policy and group-membership changes can take up to a day to take effect. Microsoft says some optimizations cut that to two hours.
An independent security write-up from Insinuator in September 2026 tested revocation timing. It found Exchange Online was consistently the slowest, sometimes taking up to five minutes even for events like password resets. That is one team's measurement, not a Microsoft guarantee, but it fits "near real time" rather than "instant."
Token Protection
This is the control aimed squarely at replay. Microsoft describes it as a Conditional Access session control that tries to reduce token replay. It requires device-bound sign-in tokens, such as Primary Refresh Tokens, which are cryptographically tied to a registered device. A stolen token shouldn't work from another machine in supported scenarios.
Its scope is narrow today, per Microsoft:
- Native applications on Windows, iOS/iPadOS and macOS are generally available.
- Browser-based support is preview only, limited to selected web apps reaching Azure Resource Manager. It isn't supported on iOS/iPadOS.
- Supported resources for native apps are Exchange Online, SharePoint Online and Teams. On Windows, they also include Azure Virtual Desktop and Windows 365.
- Supported Windows devices are Windows 10 or newer that are Entra joined, hybrid joined or registered. Windows Server 2019 or newer is supported if hybrid joined.
Microsoft itself frames Token Protection as one layer of defense in depth. It is not blanket protection against every stolen browser cookie. Its recommended rollout is a pilot group, a report-only policy, review of both interactive and non-interactive sign-in logs, then enforcement for known-compatible users.
A practical order of operations for a small team
This is my synthesis from the Microsoft guidance above, not a Microsoft-prescribed checklist.
- Require MFA broadly, and use Report-only before enforcing anything restrictive.
- Define IP-based named locations for your offices and mark the stable ones trusted. Don't add shared or unenumerable egress IPs, which Microsoft warns can weaken security.
- Use location as a step-up trigger for most users. Reserve hard blocks for countries where you never operate.
- If you own P2 licensing, add sign-in-risk and user-risk policies. These put the travel, token and AiTM detections to work.
- Prefer compliant or hybrid-joined device requirements over IP lists, especially with VPNs or proxies.
- Pilot Token Protection on Windows devices for supported apps, and check sign-in logs before enforcing.
- Keep break-glass accounts excluded and tested.
The verdict
The article's thesis is sound. Identity checks should be risk-informed and should continue after login. Its framing, though, makes the tools sound smoother than they are. "Continuous" evaluation depends on app support, network topology and location type. Travel anomalies are hints to investigate. Token binding covers a defined set of platforms and apps. For a startup, the useful lesson is that the controls exist, many need P1 or P2 licensing, and they only work if you test them in report-only mode first.
References
- Microsoft Entra Conditional Access: Zero Trust Policy Engine - Microsoft Entra ID | Microsoft Learn learn.microsoft.com
- Conditional Access - Block access by location - Microsoft Entra ID | Microsoft Learn learn.microsoft.com
- Entra ID Continuous Access Evaluation unfold techcommunity.microsoft.com