ISC · Security and Confidentiality and Privacy · 6 practice questions
AiTM vs. OAuth Consent Phishing: How to Tell Them Apart
A session captured during a reverse-proxied login points to AiTM phishing; malicious app permissions granted through employee consent point to OAuth consent phishing. Below, compare the same mailbox incident with only the access-path evidence changed.
Try one first
Hint
Focus on what the attacker was still able to use after the password changed.
Answer C. The facts point to an adversary-in-the-middle phishing attack that captured a live authenticated session after the users completed MFA. The key clue is that the attacker remained active even after password resets, which indicates session cookies or tokens, not just credentials, were stolen. The best treatment is to revoke active sessions or tokens immediately and force reauthentication, with phishing-resistant MFA as the stronger longer-term control.
Why not A: This is tempting because multiple users were affected, which can make candidates think of a broad credential attack. However, password spraying usually shows repeated failed login attempts across accounts, and the stem specifically says that pattern was not present. It also does not explain why attacker access persisted after password resets.
Why not B: This is tempting because stolen credentials are involved and breached-password screening is a real control improvement. But credential stuffing is the reuse of previously compromised username-password pairs, not a live phishing flow where the victim completes MFA on a fake page. Here, the attacker effectively bypassed the protection value of the current MFA implementation by stealing the authenticated session.
Why not D: This is tempting because password resets and mailbox cleanup are common phishing responses. The problem is that this attack did not stop at password theft; it involved session takeover, so waiting for sessions to expire leaves an active access path in place. The treatment is incomplete unless current sessions or tokens are explicitly revoked.
Side by side
| AiTM phishing | OAuth consent phishing | |
|---|---|---|
| Same mailbox incident | An employee clicks an email link; unauthorized mailbox access follows. | An employee clicks an email link; unauthorized mailbox access follows. |
| Login path | A phishing page proxies the real identity provider's login. | The employee signs in on the genuine identity provider's page. |
| Employee action | Enters credentials and completes MFA during the proxied login. | Approves a third-party app's delegated permissions to read and send mail. |
| What the attacker obtains | The authenticated session created during that login. | An app grant and issued tokens for delegated API access. |
| Mailbox access evidence | The attacker uses the captured session from another location. | Cloud API logs attribute mailbox activity to the employee-approved app. |
| Decisive classification clue | Session capture during a reverse-proxied login, not session theft alone. | Employee consent followed by the app's use of delegated permissions. |
Key points
- Remember: A session captured during a reverse-proxied login signals AiTM; employee-approved delegated app permissions signal OAuth consent phishing.
- A genuine sign-in page does not make the third-party app's requested permissions safe.
- An email link starts both incidents; the access evidence determines the classification.
How the exam traps you
- Treating any stolen authenticated session as proof of AiTM phishing. Look for evidence that the phishing page proxied the real login and captured the resulting session.
- Assuming the email link means the attacker stole the employee's password. Check whether an employee-approved app used delegated permissions to access mail through APIs.
Question 2
Hint
Which missing control would allow an unchanged, valid message to be accepted again?
Answer B. The critical fact is an unchanged, previously valid signed message was captured and later retransmitted and accepted, the classic replay attack. Acceptance occurred because the portal lacked per-request freshness (unique nonces/timestamps); adding those controls prevented reuse. Forensic findings excluding signature compromise, message alteration, or credential theft confirm replay as the primary threat.
Why not A: This is tempting because interception explains how the attacker obtained the message, but interception alone does not explain why the portal accepted the same unchanged message later. The decisive factor is the absence of freshness controls that allowed replay.
Why not C: This distractor appeals because valid signatures suggest legitimate authentication, but impersonation requires stolen credentials or signature forgery. The stem states the signing algorithm was secure and there is no evidence of credential theft; the attacker simply resent the original signed message.
Why not D: Tampering is attractive because it directly creates unauthorized changes, but forensic review explicitly found the request was not altered. With no modification, the incident is best classified as replay rather than tampering.
Question 3
Hint
Identify the attacker's live access mechanism and choose the action that most directly shuts it off first.
Answer D. The observable chain, MFA fatigue (multiple denied pushes then one approved), mailbox forwarding, and immediate OAuth app downloads, points to an active identity-based compromise and malicious app consent. Immediate containment should cut off the attacker's live access paths (disable the account, revoke sessions and tokens) while preserving mail and cloud audit logs for investigation. Network blocking or endpoint rebuilds can follow, but they risk delaying the most effective, targeted containment.
Why not A: This is tempting because network blocking seems quick, but identity-based attacks using valid credentials or OAuth tokens are not reliably stopped by blocking one or more IPs. The attacker may reconnect from another IP or maintain access through issued tokens and forwarding rules, so IP blocking alone is insufficient.
Why not B: A broad reset may be appropriate later, but it would delay stopping the currently active compromise tied to a specific user and app. The immediate priority is to stop the known attacker access on the compromised identity rather than performing organizationwide remediation first.
Why not C: Endpoint compromise is a valid concern, but the scenario shows cloud-side abuse with no endpoint alerts and explicit malicious OAuth consent. Reimaging before cutting off the attacker delays stopping ongoing exfiltration; account- and token-level containment should come first, with endpoint forensics/rebuilds afterward if needed.
Question 4
Hint
Focus on what the attacker needed the victim’s browser to do after the victim was already authenticated, and separate that from attacks that require code injection, cookie theft, or retransmission of a captured message.
Answer D. CSRF happens when an attacker induces a victim's already-authenticated browser to send unintended requests to a trusted site. Here the requests originated from the employees' browsers while they were logged in, and the portal lacked anti-CSRF token validation, which makes forged in-browser requests feasible. The absence of stolen cookies, injected portal code, or captured-and-replayed requests further supports CSRF as the best-supported classification.
Why not A: Reflected XSS requires execution of attacker-supplied script in the context of the target application (e.g., via a vulnerable page that reflects input). The investigation found no evidence of code injected into or stored on the treasury portal, making XSS unlikely here.
Why not B: Session hijacking implies the attacker took over the user's session, typically by stealing or replaying the session identifier. The facts indicate no evidence that session cookies were stolen; instead, the users' own browsers submitted the unauthorized requests, which is consistent with CSRF rather than hijacking.
Why not C: A replay attack depends on capturing and resending previously valid requests or credentials. The stem explicitly notes there was no indication that prior valid requests were captured and resent, so replay is not supported by the evidence.
Question 5
Hint
Focus on identity provider and cloud API logs, did the activity use a delegated token from an app consent or direct password/MFA authentication?
Answer A. IdP logs show an OAuth access token issued after a user consent event and cloud API logs attribute the downloads to that third-party app using the token. The absence of password/MFA authentication events or endpoint malware makes a delegated OAuth-consent compromise the most consistent explanation: the attacker gained API-level access via the user's consented app rather than by stealing credentials or implanting malware.
Why not B: Tempting because the user clicked a phishing link and exfiltration followed, which often indicates credential theft. It's unlikely here because logs show an OAuth token issued to a third-party app and the downloads were performed by that app via cloud APIs rather than by an interactive user session authenticated with the manager's password.
Why not C: Plausible because no malware was found and the access resembles legitimate API activity. However, forensic and IdP/API logs specifically attribute the downloads to a third-party application using an access token issued after user consent, not to an interactive user session or a manual file transfer from the manager's workstation.
Why not D: Advanced backdoors can exfiltrate data stealthily and sometimes leave little trace, making this tempting. In this case, endpoint scans found no malware and cloud API logs show the third-party app (using an OAuth token) performed the downloads, evidence inconsistent with a workstation-resident backdoor funneling data to a C2 server.
Question 6
Hint
Identify what actually gave the attacker durable access: guessed credentials, a hijacked session, or a delegated permission grant.
Answer B. Users explicitly approved a third-party application that received delegated permissions to read and send mail and to retain access, which explains the API-based activity and why revoking the app stopped the attacker. These facts point to malicious OAuth consent (consent phishing) rather than stolen credentials or cookie replay.
Why not A: This option is tempting because the attack began with a targeted email and a sign-in prompt, but credential harvesting would normally result in direct interactive sign-ins or credential-replay indicators. The scenario instead describes an app grant enabling delegated API access.
Why not C: Password spraying is a brute-force technique that typically generates many failed authentication attempts across accounts. The provided facts show successful user consent and API access without the failed-sign-in patterns characteristic of password spraying.
Why not D: Session hijacking via stolen cookies can permit access without credentials, but that attack usually produces evidence of token reuse from unusual locations or sessions. The decisive evidence here is the app consent and the fact that revoking the application's grant ended the activity, which aligns with delegated OAuth access rather than cookie theft.
Common questions
How do I distinguish AiTM from OAuth consent phishing?
AiTM captures a session during a reverse-proxied login. OAuth consent phishing gets an employee to grant a malicious app delegated permissions, which the app uses for API access.
Can OAuth consent phishing use a genuine sign-in page?
Yes. The employee can authenticate on the genuine identity provider's page, then approve a malicious third-party app. The app grant, not a fake password form, explains the mailbox access.
Does a stolen session always mean AiTM phishing?
No. Session theft alone does not establish how the attacker obtained it. In this comparison, the reverse-proxied login is the distinguishing evidence for AiTM.
Watch it solved
A full CPA ISC task-based simulation on Threats and attacks, worked step by step.
SIEM Attack Identification | CPA ISC Simulation (TBS) on YouTube
Practice ISC like the real exam
The free ChatCPA simulator: real exam layout, timed testlets, starting with a question on this topic. No account needed to start.
Open the free simulator →More on Threats and attacks
Questions from the ChatCPA bank of 17,658 CPA exam questions, each with a written reason for every wrong answer. ChatCPA is built by Nicholas Miller, CPA (Oregon #14907). How these pages are made. Spot an error? Tell us.