Skip to content

LearnSocial engineering

Help desk identity verification testing

Brief

A help desk reset is the softest way around MFA, because the person on the phone decides who you are from what you tell them. This page tests that decision with a script the client approved and an identity the client created for the test.

Stage
Access
Reviewed
On this page
  1. Check first
  2. How do I set up the test identity and the approved script?
  3. How do I test a password reset request over the phone?
  4. How do I test a callback to a number on file?
  5. How do I test an MFA reset or re-enrolment request?
  6. How do I keep the recordings and the call log lawful?
  7. The defender sees
  8. Done when
  9. Not covered
  10. Related research
  11. Reviewed

All commands

Situation
The scope letter allows calls to the help desk and names the numbers, the hours and the number of calls. The client has created a test identity in the directory with a manager, a phone number on file and a mailbox we can read, and has approved the script word for word. Calls are recorded only where the scope letter says so and the law at both ends of the call allows it. No real employee's account is touched.
Goal
A call log: each call, the verification steps asked, which ones we passed with the agreed public footprint, and whether a reset happened.
Outputs
Representative examples on placeholder hosts, trimmed to the lines that matter. Not captured from a public fixture yet.

Check first

  • The help desk's written verification procedure, if one exists, so the test measures it against its own rules as well as ours.
  • The test identity: created by the client, with a manager, numbers on file, a start date and a mailbox we can read.
  • The agreed public footprint for that identity: name, title, team, office, manager's name, start month. We answer with nothing beyond it.
  • The ticketing and phone systems the help desk uses, and whether agents are told a test is possible this quarter.
  • Recording law at both ends of the call; where unclear, we take notes instead.
  • Who resets the test identity afterwards, and when.

Steps

How do I set up the test identity and the approved script?

The client approves every line of the script. We present as the test identity, never as a real person, ask for one thing per call, and answer only from the agreed footprint. Asked for anything outside it, we say we do not have it and let the call end as it would for a stranger.

Identity:  created by the client in the directory, manager set, numbers on file,
           standard user with no group memberships, noted by the SOC as a test account
Calls:     1 password reset, 1 callback request, 1 MFA re-enrolment, each once
Answers:   name, title, team, office, manager, start month; nothing else, ever
Stop:      any sign of distress on the line, any mention of a real colleague's account

Expected output

Identity record:  <test identity>   Manager: set   Mobile on file: set   Mailbox: readable
Footprint check:  6 attributes agreed, 0 secrets included   # <- the identity can only be verified by something we were not given

Negative result. A test identity with no manager or number on file cannot test the callback or approval steps; the client completes the record first.

Fix. Verification depends on data: the joiner process records a manager and a verified phone number for every account on day one, since an agent cannot call back a number that was never recorded.

How do I test a password reset request over the phone?

We call during the agreed hours, ask for a password reset for the test identity, and log every question asked and whether the footprint could answer it. The call ends when the agent resets or refuses.

Request:  "I am locked out and need my password reset."
Answer:   name, title, team, office, manager, start month, as the footprint gives them
Decline:  "I do not have that" for anything else, then let the call run its course
End:      a reset, a refusal, or the stop condition

Expected output

Call 1  10:04  password reset requested
  asked: full name, manager, office            answered from footprint
  asked: employee id                           answered from footprint (printed on the badge the client described)
  reset: performed, new password read out on the call   # <- every check was answerable by anyone who read the staff directory
  ticket: INC0041 opened, verification field blank

The client changes the password within the hour and we never sign in with it; the reset is the finding.

Negative result. The agent sends a one-time code to the number on file, or asks for a check only the real person could complete, refuses when we cannot, and records the refusal in the ticket.

Fix. Checks that cannot be answered from public information: a code to the number or device already on file, a video check against an identity document for high-risk roles, or an in-person visit. Date of birth, employee id, manager's name and start date are not secrets.

How do I test a callback to a number on file?

The second call asks for the same reset, says the number on file is out of reach and offers a new one.

Request:  the same reset
Statement: "My work phone is in the office, can you use this number instead?"
Offer:    one callback number we control, given once; no insistence if refused
End:      a callback to the new number, a refusal, or the stop condition

Expected output

Call 2  10:31  reset requested, new callback number offered
  agent: "I can only call the number we have on record"   # <- the control held; the call ends without a reset
  ticket: INC0042 closed, reason recorded: caller could not be reached on file number

Negative result. The agent calls the number on file and nothing else; a number change goes to a separate, verified process.

Fix. Callbacks go to the number already on file, with no exception an agent can grant on the phone. A number change is a separate request with manager approval, since whoever can change the number first has defeated the callback.

How do I test an MFA reset or re-enrolment request?

The third call asks for the test identity's authentication methods to be cleared so a new phone can register. This is the request that ends real engagements with a takeover, as the MFA field note and the token replay analysis describe.

Request:  "I have a new phone and the authenticator app is gone, can you clear it so I can set it up again?"
Answer:   footprint only; the manager's name if asked, never a number for the manager
Stop:     the agent asks us to supply a contact for approval; we decline and end the call
End:      methods cleared, a refusal, or an approval request to the manager on record

Expected output

Call 3  11:02  MFA re-enrolment requested
  agent: manager approval required, manager contacted on the number in the directory
  manager (client, briefed): declined
  audit log: no "Admin deleted security info" entry for the test identity   # <- the methods were never touched
  ticket: INC0043 closed, approval refused

Negative result. No change to the test identity's methods, and a ticket showing who approved and how they were reached.

Fix. MFA resets need approval from the manager on record, reached through the manager's directory number and never through a contact the caller supplies. Our recommendation for an approved reset is a time-limited credential such as a Temporary Access Pass, rather than clearing methods and leaving the account on a password alone.

How do I keep the recordings and the call log lawful?

Recording is agreed in the scope letter or it does not happen. Where agreed, we announce it where the law requires, store the files encrypted, share them only with the named contact, and delete them when the report is accepted. The log names agents by shift, never by name, and every failed check is written as a gap in the procedure.

Recording:  only where the scope letter allows it and the law at both ends does; announced where required
Storage:    encrypted, shared with the named contact only, deleted when the report is accepted
Log entry:  time, request, checks asked, checks passable from the footprint, outcome, ticket id
Names:      agents by shift, test identity by its label, no real employee anywhere

Expected output

Retention:  recordings deleted 2026-11-10, confirmed to the client in writing
Log fields: time, request, checks asked, checks passable from footprint, outcome, ticket id   # <- the record a defender can match to its own

Negative result. Every call maps to a ticket the client can reconcile against its phone records.

Fix. A required verification field on the ticket: which check was performed and how the caller passed it. An agent who cannot fill it cannot close the ticket, and a reviewer can audit resets without listening to calls.

The defender sees

Each call is a ticket and a call detail record in the phone system, which shows the caller's number against the number on file. In Microsoft Entra ID, a reset performed by the agent appears in the audit log as Reset password or Reset password (by admin), and a cleared or added method as Admin deleted security info or Admin registered security info, with the agent as actor and the test identity as target. In Active Directory, the domain controller logs 4724 for a reset of another account, agent as Subject and test identity as Target; 4723 is a user changing their own password and should not appear. A reset with no ticket, an empty verification field, or an MFA change with no approval recorded is each a finding. The report says whether anyone reviews those events and whether a reset of a privileged account would raise an alert.

Done when

Every planned call is made once, each reset or refusal is matched to the ticket and the directory audit log, and the client has reset the test identity's password and methods afterwards.

Not covered

Calls to real employees, pretexts that threaten or claim authority, chat and email channels, and the directory itself; every output is representative on a placeholder tenant.