An employee receives a document invitation, follows the sign-in instructions, and lands on a genuine Microsoft page. The address looks right. The employee completes authentication successfully. The missing question is whose application receives the access that authentication approves.
Device code phishing exploits that gap. For a company assessing its Microsoft 365 exposure, checking whether employees recognize a fake login domain is only part of the problem. The assessment also needs to examine which authentication flows the organization permits and how it handles an unauthorized grant.
Understand the decision behind the code
Device authorization lets a user complete sign-in on another device for a client that cannot conveniently collect input itself. In a phishing scenario, the attacker supplies a code associated with the attacker’s client. The victim’s interaction with a legitimate identity provider can complete that client’s authorization.
Microsoft’s September 22 EvilTokens research documents campaigns that abused this flow to obtain tokens and access organizational accounts. Employees need to verify which application or device they are authorizing, even when the sign-in page belongs to Microsoft.
For example, a finance employee may receive instructions to enter a code to view a supplier’s revised invoice. That code can authorize a new client session. Before entering it, the employee should confirm the request through the supplier contact or invoice process they already use.
Find the legitimate uses before changing policy
Ask the identity administrator to identify current device-code use, the accounts involved, and the business owner for each case. Meeting-room hardware or a specific administrative tool may have a legitimate dependency. A broad exception for every employee is a different choice from an exception for a documented device group.
Microsoft’s authentication-flow documentation describes reviewing sign-in logs and using report-only Conditional Access policies to understand impact. It recommends allowing device code flow only where needed. Treat report-only results as impact evidence, not as proof that a blocking control is already active.
For a review, produce a small exception register. Each exception needs a purpose, owner, scope, and review date. Ask what prevents an attacker from using the same exception with a different client or destination resource. If the answer is unknown, keep that uncertainty in the assessment rather than treating the exception as automatically safe.
Test the policy with accounts created for the assessment
Use a test identity, synthetic mailbox content, and an approved test client. Establish a permitted baseline, then exercise a flow the policy is intended to deny. Do not ask an employee to authorize a real attacker-controlled application as a demonstration.
Keep results for both the permitted and denied cases. The legitimate device workflow should still work, and the excluded client should fail to obtain a session. A policy screenshot helps explain the setup; the test results show what happened.
This creates two meaningful outcomes: legitimate business access remains possible, and the excluded path is rejected. A screenshot of a policy’s configuration establishes intent. These results establish its behavior for the tested cases.
Rehearse containment beyond a password change
If unauthorized access occurs, the response owner needs to consider issued sessions, refresh tokens, added authentication methods, application permissions, and activity in the affected services. The correct actions depend on the evidence and the identity platform’s behavior. Do not assume resetting a password proves every issued access path has stopped.
For a tabletop exercise, create a synthetic timeline with a sign-in, mailbox access, and a suspicious rule change. Ask the team to identify which logs it needs, who can revoke access, and how it would verify containment. Check the retention period before an incident, since a missing log cannot be recovered by writing a better response procedure later.
Measure the handoff between help desk, identity administrator, and incident response. The employee’s report needs a route to someone with the authority to investigate. A technically correct control can still leave a damaging delay if every team expects another team to act.
Where this fits in a Ceron assessment
For Ceron, agree whether the engagement covers identity configuration, an authorized authentication-flow test, or the customer application that trusts the resulting identity. These scopes require different access and evidence. A website’s HTTPS or security-header results do not establish whether Microsoft 365 sign-in policies are effective.
The report should list permitted flows, exceptions, observed policy decisions, and the containment exercise results. Assign unresolved items to an owner and specify the test required to close each one.
A genuine login page can authorize access for the wrong client. Include that case in employee guidance and in the tests of your sign-in policy.
Further Ceron reading
Session revocation after compromise explains why issued access deserves separate verification. The finance and incident scenarios above are illustrative assessment designs.