Here’s the short version: Microsoft 365 app consent lets a user grant an outside app access to mail or files, and a password reset does not take that access back. The fix is a tighter consent setting, a real approval queue, and an inventory of apps that are already in.
It does not replace multi-factor authentication. MFA still matters. It does not undo a grant.
What Microsoft 365 app consent actually grants
Before an application can touch tenant data, somebody grants it permissions. Microsoft’s user consent article states the default: users may consent to permissions that do not require an administrator. The same page gives the example that matters in a small office. A user can allow an app to access their mailbox. A user cannot, by default, allow an app unfettered read and write across every file in the organization.
Delegated permissions run with a signed-in user. The app can do what that user can do, inside the scopes they approved. Application permissions run with no user present. A background app reading every mailbox is this type. Only an administrator can consent to application permissions.
User consent records access for that person. Admin consent records access for the organization. If the app is also configured so that users must be assigned to it, an administrator has to consent even when your user-consent policy would have allowed the person to approve it for themselves. A “required assignment” app is not a loophole around the queue. It is a reason the queue exists.

Why a password reset does not undo the grant
Microsoft documents this attack as an illicit consent grant. An attacker registers an app that asks for contacts, mail, or documents. A phishing message, or code dropped into a site the person already trusts, sends them to the real consent screen. After they accept, the app has account-level access without an account of its own inside the firm.
That is why the usual incident steps fail. Resetting the password changes the human’s credential. Requiring MFA challenges the human at the next sign-in. The app does not sign in as the human. It presents a token Microsoft already issued because consent said it could. Conditional Access still belongs on the human accounts. It is not a substitute for removing the grant.
Microsoft’s consent phishing article adds a second trap. If Microsoft later disables a malicious app, new token requests and refresh tokens are denied. Access tokens already issued stay valid until they expire. A tenant admin cannot undo that disablement by flipping the app back on. If your firm publishes the app and you think the ban is a mistake, that is a support case in the app’s home tenant.
A Teams phishing message can carry the same link an email would. The channel changes. The Accept button does not.
The Microsoft 365 app consent setting to change first
Microsoft’s recommendation on the user consent page is specific: allow user consent only for apps from a verified publisher. The matching built-in policy is narrower than that sentence. It allows apps from verified publishers and apps registered in your tenant, and only for permissions you classify as low impact.
| Portal choice | What a user can do | When we use it |
|---|---|---|
| Allow user consent for apps | Any permission that does not already require an admin, for any app | Almost never. This is the wide default. Leave it only if you have a written reason. |
| Allow user consent for apps from verified publishers, for selected permissions | Verified publishers and your own apps, and only permissions you marked low impact | This is the setting for a small firm. |
| Do not allow user consent | Nothing. Every app waits for an admin. | Only with the request workflow on. A silent ban trains people to find a workaround. |
If the tenant is still on the wide legacy choice, that is the first change. A total ban with no request path pushes people toward personal accounts and files sent out so a tool still “works.”
Verified publisher is a badge, not a permission review
Publisher verification is Microsoft’s vetting of the developer. A badge does not replace reading the prompt. Attackers copy product names and domains. Spelling and grammar are a reason to stop, not a complete test. The prompt has a Report it here link. Microsoft reviews flagged apps and can disable one that breaks the terms. A verified app that asks to read and send mail, when the person only wanted a scheduling link, is still the wrong yes.
Classify the low-impact list or the policy is empty
The verified-publisher policy only helps if you classify permissions. Microsoft’s configure page says the low-impact policy allows the permissions you mark as low impact. If you flip the setting and never classify anything, you have not chosen what “low” means.
The investigation playbook puts sign-in basics at the low end: User.Read, User.ReadBasic.All, openid, email, and profile. offline_access belongs there only beside those. It is not a free pass next to Mail.Read.
Anything that reads or sends mail, reads files, or changes directory data stays off the user list. Those go through the queue.
Turn the admin request queue on the same day
If users cannot consent, they need a way to ask. The admin consent workflow is that way. You turn on “Users can request admin consent to apps they are unable to consent to,” you name reviewers, and you decide whether those reviewers get email. Microsoft says the workflow can take up to an hour to start working. A test in the first five minutes is not a failed rollout.
A few rules surprise people:
- A reviewer can see requests, and can block or deny them. Being a reviewer does not grant them the right to approve every permission.
- Only a Global Administrator can approve admin consent when the app wants Microsoft Graph application permissions, the app-only roles.
- A reviewer added later cannot act on requests that already exist or have already expired. If you turn off reviewer email, requesters stop getting email too.
Global Admin hygiene still applies. Do not hand that role to the help desk so they can clear the queue. Approving application permissions stays a Global Admin act.
One approval can cover every mailbox
Microsoft’s review article says that approving a request lets all users in the tenant use the app, unless you also require user assignment. The person who asked may have wanted a pilot. The approval can cover accounting, the partners, and the shared mailboxes.
When the permissions are wide, grant consent and require assignment to a group. The pilot group gets the app. The rest of the firm does not inherit it.
Risky consent has a separate brake. Microsoft’s risk-based step-up is on by default, and it only changes the prompt while user consent is still allowed. A risky app tells the user an admin must approve. With the workflow on, they can send that request from the prompt. With it off, they get an error and no queue.
Permissions that show up in consent phishing
Microsoft’s app consent investigation playbook lists the delegated permission families its incident response team sees misused. The first six show up, in combination, in 99% of the consent phishing attacks they describe:
Mail, including send, but notMail.ReadBasicContactsMailboxSettingsPeopleFilesNotes
Directory.AccessAsUser.All and the blanket User_Impersonation scope are on the same caution list. Application-permission versions of mail, files, and directory write scopes are worse, and a normal user cannot approve those. The attacks still aim at users, because a delegated Mail.Read or Files.Read on one partner mailbox is enough.
The same playbook says to look for consent type AllPrincipals. That value means an administrator consented for the whole tenant. Microsoft’s own apps often need it. Every non-Microsoft app with it deserves a name, an owner, and a reason.
Copilot oversharing is a different door. Copilot reads what a person can already open. A consented app reads what the grant allows. Cleaning SharePoint links does not revoke the app. Revoking the app does not fix an “Anyone” link.
How you find an app that is already approved
Do this before you argue about the setting. New policy does not rewind grants that are already there.
Microsoft’s remediation article gives three ways to see them. In the Entra admin center, open a user and then Applications. Users can open myapps.microsoft.com and revoke an app they do not recognize. The faster tenant-wide view is the permissions export that article points at.
Then search the audit log for Consent to application. A value of true on the admin-consent detail means someone with Global Administrator access may have granted broad access. Treat that as a lead. Confirm the app, the permissions, and who consented. You will not reconstruct what the app read if mailbox auditing and user activity auditing were off at the time. Microsoft says so on that page. A dark log is not a clean bill of health.

While you read the list, separate three piles:
- Microsoft and the vendors you already pay, with permissions that match the product.
- Unknown names, misspellings, or apps whose only story is “someone needed it once.”
- Non-Microsoft apps with tenant-wide consent, especially mail, files, or directory write.
How you remove a bad grant without breaking the tenant
When the app is illicit, Microsoft’s remediation article lists the moves that remove access. Remove the application on the user’s Applications page. Revoke the OAuth permission grant. If an admin consented to an app role, revoke that assignment too. None of those is “change the password and hope.”
Disabling sign-in on the affected account stops the app from reaching that mailbox, and it stops the person from working. Use it as a short brake while you revoke the grant.
Turning off integrated applications for the whole organization is the drastic switch. Microsoft’s article says not to recommend it. It blocks the malicious prompt and every other non-Microsoft app the firm uses. If you need a few quiet hours during an incident, do it on purpose and turn a reviewed path back on. Do not leave the tenant with no apps and no queue.
We do not delete a service principal that Microsoft disabled for a terms violation. The consent phishing article says those objects stay so the app cannot be created again, and you cannot simply re-enable them. Security awareness training is the habit: if a prompt asks for mail or files, use the queue. Training does not replace the inventory.
What we will change, and what we will not ban
On a tenant we manage, the week-one app consent work looks like this.
- Export or otherwise list consented apps. Note tenant-wide grants and anything with mail, files, or directory write.
- Move user consent off the legacy “any app, any non-admin permission” setting. Use verified publishers and low-impact permissions, after those permissions are actually classified.
- Turn on the admin consent workflow the same day. Name two reviewers. Keep Global Admin approval for application permissions.
- Require assignment on any admin-consented app that does not need to be universal.
- Remove grants that have no owner. Do not disable every integrated app to feel done.
- Tell staff, in one short note, that Accept is a signature and that the request form is the path. Then stop sending notes and start reviewing the queue.
We will not call a verified badge a security review, and we will not tell you MFA closed an app that is still holding a token. If the queue and the bad grants should be someone else’s weekly job, start with cybersecurity and contact Secure Techies. The Accept button is already in the tenant. The setting decides who gets to press it next.
