
Most Entra ID attack paths to Global Admin run through legitimate features, not exploits: a phished token, an over-privileged app or group, and a hybrid sync server treated as ordinary infrastructure. Each hop has a concrete control that removes it, and most tenants we assess leave at least one of them open.
The short answer
- Entra ID attack paths reach Global Admin by chaining a phished token to apps or groups that already hold privileged roles.
- Device code and consent phishing yield valid tokens after the victim completes MFA.
- Block device code flow, restrict user consent, keep admins cloud-only and treat app and group owners as privileged.
Key takeaways
- Entra ID attack paths to Global Admin chain legitimate features: phished tokens, app ownership, role-assignable groups and hybrid sync.
- Device code and consent phishing produce valid tokens after the victim completes MFA, so MFA alone does not stop them.
- Anyone who can add credentials to an app with high Graph permissions effectively holds Global Admin.
- Owners of role-assignable groups, and the Entra Connect server, belong in Tier 0.
- Blocking device code flow, restricting user consent and keeping admins cloud-only removes the most common hops.
What is an Entra ID attack path from a phished user to Global Admin?
An Entra ID attack path is a chain of trust relationships that lets a low-privilege identity reach a directory role such as Global Administrator. The first hop is usually a token, not a password. The later hops are ownership and assignment edges: who owns which app, who can add members to which group, and which server can write to the directory.
The pattern we see in assessments has three stages:
- Initial token. Device code phishing or an illicit OAuth consent grant gives the attacker a valid session or refresh token for a real user. The victim completed MFA, so MFA does not help.
- Privilege edge. That user, or an app they control, connects to something more privileged: an app registration with high Microsoft Graph permissions, a role-assignable group, or a delegated admin role scoped wider than intended.
- Directory control. The privileged object grants a directory role, or a hybrid component (Entra Connect, AD FS, pass-through authentication) turns on-premises compromise into cloud admin.
None of these steps is unusual. Microsoft’s own hardening guidance covers each one. The trouble is that different teams own them: identity owns Conditional Access, app teams own registrations, infrastructure owns the sync server. Nobody owns the path.

Why do device code phishing and consent phishing get past MFA?
Both techniques end the same way: the victim completes MFA on a genuine Microsoft page, and the attacker walks away with a valid token. MFA proves the right person approved the sign-in. It says nothing about who ends up holding the result.
Device code phishing
The OAuth device authorization flow exists so devices without a browser (TVs, CLI tools, conference room hardware) can sign in. The attacker starts the flow, receives a short user code and persuades the victim to enter it on the real Microsoft sign-in page. The victim sees a real domain and their usual MFA prompt.
MFA is not bypassed here. The legitimate user completes it, for the attacker. The tokens go to the session the attacker started, not to the victim’s browser.
Two details make this worse for defenders:
- The refresh token can often be redeemed for access tokens to other first-party resources, so a phish “for email” becomes access to Teams, SharePoint and Microsoft Graph.
- Sign-in logs record a successful interactive sign-in with MFA satisfied. Unless a rule checks the authentication protocol field, it looks normal.
Most organizations need device code flow only for a small set of devices and accounts, which makes it one of the cheapest hops to remove.
Illicit OAuth consent grants
An illicit consent grant happens when a user approves a malicious multi-tenant app that requests delegated permissions to their data. After consent, the app holds its own tokens and never needs the user’s password or MFA again.
The scopes look mundane: read mail, read files, offline_access. The last one matters most because it yields a refresh token. Resetting the user’s password does not revoke a consent grant.
Consent phishing turns into privilege escalation in two cases:
- The consenting user is an admin. An administrator can grant tenant-wide consent, including application permissions that act without any user, such as reading every mailbox in the tenant.
- The tenant allows broad user consent. If any user can consent to any app, one click creates a durable service principal in your tenant that you have to find and remove.
Watch for a new service principal followed by a consent event, especially for an app from an unverified publisher with a recently created home tenant.
Can a standard user become Global Admin in Entra ID?
Yes, but almost never directly. A standard user reaches Global Admin by sitting one or two edges away from an object that already holds that power. Entra ID privilege escalation is mostly unreviewed ownership and delegation.
The edges we find most often in real tenants:
- A regular user listed as owner of an app registration with high Microsoft Graph application permissions.
- A regular user who owns a role-assignable group carrying a privileged directory role.
- Helpdesk staff given Privileged Authentication Administrator, which can reset credentials and MFA for any user including Global Admins, because a narrower role could not close certain tickets.
- Developers holding Application Administrator or Cloud Application Administrator just to make things easier.
Each edge is legitimate in isolation. Drawn as a graph, they connect a phishable mailbox to directory control. That graph is what an Entra ID security assessment should produce, not a list of misconfigured toggles.
App registrations and service principals
App registration abuse works because a service principal can hold permissions no human uses day to day, and whoever manages its credentials inherits them. Application permissions that manage role assignments, manage all applications or write to the directory amount to Global Admin in practice, even on an app labeled as a reporting tool.
Who can add a client secret, certificate or federated credential to such an app?
- Owners of the app registration or service principal.
- Holders of Application Administrator or Cloud Application Administrator, tenant-wide.
- Any other service principal with permission to manage applications.
A compromised developer or app admin can attach a new credential to a privileged app and act as it. The app signs in as a workload identity, so user-focused Conditional Access does not apply unless you add workload identity policies.
Review federated credentials separately. Through workload identity federation, an external identity provider (a CI system, another cloud) exchanges its own token for an Entra access token for the app. A federated credential that trusts an issuer or subject you do not control is a backdoor with no secret to rotate.
Role-assignable groups
Role-assignable groups are security groups that can receive Entra directory roles, so every member inherits the role. Membership becomes role assignment, and ownership becomes the ability to grant the role.
Microsoft added guardrails. Only Privileged Role Administrator or Global Administrator can create these groups. The isAssignableToRole property is set at creation and cannot be changed afterward. Only Privileged Authentication Administrator or Global Administrator can reset credentials of members; Helpdesk and Password Administrators cannot. None of that helps if ownership is dirty.
Failure patterns we see:
- Owners who are ordinary users, often whoever requested the group years ago.
- Global Administrator or Privileged Role Administrator assigned to the group permanently rather than as eligible through Privileged Identity Management (PIM).
- Access packages or self-service join settings copied over from a different group.
Treat every owner of a role-assignable group as holding the role itself. If that changes who you consider privileged, the review has done its job.
How does Entra Connect or AD FS create a hybrid path to Global Admin?
A hybrid deployment turns on-premises compromise into cloud compromise, because Entra ID trusts the bridging components to vouch for identities.
- Entra Connect server. The sync service holds credentials for an account that writes to Entra ID. Control of that server is control of directory writes. It belongs in Tier 0 with your domain controllers.
- Pass-through authentication agents. On-premises agents validate passwords for cloud sign-ins. A compromised agent host can influence those decisions.
- AD FS. The token-signing certificate lets its holder mint SAML tokens that Entra ID accepts for any federated user, including admins, without an MFA prompt.
- Desktop SSO computer account. Single sign-on for domain-joined desktops relies on a Kerberos key held by the AZUREADSSOACC computer account. Microsoft recommends rotating it regularly; in many tenants it has not changed since setup.
The deciding control is cloud-only Global Admins. If privileged cloud accounts sync from Active Directory, a domain compromise reaches Global Admin by design. This is where Active Directory penetration testing and cloud testing have to share scope, because the path crosses the boundary.

The controls that break each hop
Conditional Access with authentication flow controls breaks device code phishing. Restricted user consent with an admin consent workflow breaks consent phishing. The rest of the chain needs its own controls.
| Attack hop | Control that removes it | What to verify |
|---|---|---|
| Device code phishing | Conditional Access policy blocking device code flow, with a scoped exclusion list | Device code sign-ins fail for standard users |
| Illicit consent grant | User consent limited to verified publishers and low-risk scopes; admin consent workflow | A user cannot consent to a test app from an unverified publisher |
| Token replay | Token protection, compliant or hybrid-joined device requirement, continuous access evaluation | On supported clients and resources only, a token replayed from another device is rejected |
| Push or SMS phishing of admins | Authentication strength requiring phishing-resistant methods (FIDO2, Windows Hello, certificate) | Admin roles cannot sign in with push or SMS |
| App credential abuse | Limit Application Administrator; app instance property lock; workload identity Conditional Access | No non-admin owners on apps with high Graph permissions |
| Role-assignable group takeover | PIM for groups; eligible, not permanent; owners only from the admin tier | Owner list per role-assignable group matches the admin roster |
| Hybrid pivot | Cloud-only admins; Entra Connect and AD FS in Tier 0; move from AD FS to cloud authentication | No privileged role holder is a synced account |
Conditional Access and MFA gate the sign-in event; they do not stop token theft. Once a token exists, token protection, device compliance and continuous access evaluation limit where and how long it works. Token protection coverage varies by client and resource, so test it against the apps you actually use.
Which Entra ID roles to protect first
Start with the roles that grant other roles or take over credentials, not the ones with the scariest names.
- Global Administrator. Fewer than five, cloud-only, phishing-resistant MFA, eligible through PIM, plus two monitored break-glass accounts excluded from Conditional Access.
- Privileged Role Administrator. Can assign any role, including Global Admin.
- Privileged Authentication Administrator. Can reset credentials and MFA for any user, including Global Admins.
- Application Administrator and Cloud Application Administrator. Can add credentials to privileged apps; scope them to specific apps where possible.
- Hybrid Identity Administrator and whoever administers Entra Connect hosts. They control the bridge.
Service principals with equivalent Graph application permissions belong on the same list. A role audit that only lists humans misses them.
How do I detect malicious OAuth apps and app credential changes?
Malicious OAuth app detection comes down to alerting on a short list of Entra audit events and reviewing each one with context: who did it, and to which app.
- Consent to application, particularly with offline_access or mail and file scopes, or from an unverified publisher.
- Add service principal for an app whose home tenant is not yours.
- Update application or service principal with a new certificate, secret or federated credential, especially on apps with high Graph permissions.
- Add owner to application or service principal.
- Add member to a role-assignable group, and any direct role assignment outside PIM.
In sign-in logs, filter the authentication protocol field for device code sign-ins and compare against your allowed list. For workload identities, flag service principal sign-ins from new IP ranges or to first-seen resources. A credential added to a dormant app on a Friday evening should page someone that night, not surface in a post-incident report.
Run a quarterly inventory: every app with application permissions, its owners, its credential expiry dates and whether anyone can explain why it exists. If OAuth and OIDC security testing covers your own apps, include consent and redirect handling there too.
How do you remediate a compromised Entra ID app or user token?
Remediation means revoking every credential the attacker could hold, not just the password, and removing anything they added.
For a compromised user:
- Revoke sessions and refresh tokens, then reset the password and re-register MFA methods, checking first for attacker-added methods.
- Remove consent grants the user made during the incident window.
- Check for new inbox rules, app registrations, owned objects and devices registered to the account.
For a compromised app or service principal:
- Remove every secret, certificate and federated credential, not only the one you know about, then issue a new credential through a controlled change.
- Remove unexpected owners and review permission grants for changes.
- Disable the service principal while you investigate its sign-in activity, if the business can tolerate it.
Then walk the path backwards. If the attacker reached an app through an owner relationship or a group, that edge survives cleanup unless you remove it. Validate the fix the way the path was found: by retesting the specific chain, not by rereading a configuration checklist.

FAQ
Does MFA stop device code phishing in Entra ID?
No. The victim completes MFA on the genuine Microsoft page, and the tokens go to the session the attacker started. A Conditional Access policy that blocks device code flow, with a narrow exclusion list, is the effective control.
Does a password reset remove an illicit OAuth consent grant?
No. The malicious app keeps its own refresh token, so a password reset leaves its access intact. Revoke the consent grant, remove the service principal, then revoke the user's sessions.
What is a role-assignable group in Entra ID?
It is a security group that can receive directory roles, so every member inherits them. Its owners can add members and effectively grant that role, which makes group ownership a privileged position.
Should Global Admins be synced from on-premises Active Directory?
No. Synced admin accounts let a domain compromise reach the cloud tenant. Use cloud-only accounts with phishing-resistant MFA for every privileged Entra role.
Do Conditional Access policies apply to service principals?
Standard user policies do not. Workload identity Conditional Access can restrict service principal sign-ins by location and risk, and it requires the appropriate license.
How many Global Administrators should a tenant have?
Microsoft recommends fewer than five, plus two monitored break-glass accounts. Day-to-day work should use narrower roles activated through Privileged Identity Management.
agent::offense maps the full chain from a phishable mailbox to directory control and shows which edge to cut first. See our Entra ID security assessment.