Types of Phishing Attacks and Fraud: SOC Analyst Cheat Sheet
Types of phishing attacks and fraud, from quishing to BEC: what the victim sees, the artifact each one leaves in your logs and the first check to run.
21 min read
On this page
- The cheat sheet: types of phishing attacks and fraud
- Stage 1: Lure
- Stage 2: Access
- Stage 3: Money
- Types by how the lure is built
- The three stages of every fraud
- Stage 1: The lure
- Stage 2: The access
- Stage 3: The money
- Three schemes that get misread
- Quishing: the page is the payload, not the code
- AiTM: continuity is the signal
- Invoice fraud: a procedure stops it, not a tool
- The rule that hides the fraud: one query
- The logs to pull, by stage
- The first ten minutes
- Five pairs that get confused
- Map the schemes to ATT&CK
- A worked example: the invoice that arrived on a Thursday
- What this cheat sheet does not cover
- Pick your first operation
- Not every phish has a link to click
Fraud reaches a SOC as a message, a sign-in or a payment request, and rarely as a malware alert. This cheat sheet covers the types of phishing attacks and fraud that show up in that gap: email phishing, spear phishing, whaling and clone phishing by how the lure is built, then quishing, smishing, credential harvesting, MFA fatigue, AiTM, vishing, BEC and invoice fraud by what they leave behind. For each one: what the victim sees, the artifact in your logs and the first check that tells you which one you have.
The FBI's IC3 recorded $20.9 billion in reported losses in 2025. Business email compromise accounted for $3.05 billion of that across 24,768 complaints, about 15% of the total. Phishing and spoofing was the most-reported category, with 191,561 complaints.
Read the cheat sheet first. The sections after it cover the schemes analysts misread most often, one query for the artifact that fraud keeps hiding in, the logs to pull by stage and the first ten minutes of a live report. Each scheme links to an operation where you can work it from the logs.
The cheat sheet: types of phishing attacks and fraud
The schemes run in stage order: lure, access and money. Victim sees is the story the attacker tells and the only part users report. SOC sees is the artifact that survives the story. First move is the one check that separates this scheme from the other seven.
Stage 1: Lure
| Scheme | Victim sees | SOC sees | First move | Practice |
|---|---|---|---|---|
| Quishing | A QR code on a parking meter, a table tent or a PDF invoice. | A PDF carrying an image, not a URL, so there is nothing for the gateway to rewrite. The scan arrives from a mobile carrier's ASN (the network that owns the IP address), outside the proxy. | Decode the QR in a sandbox and resolve the final URL, not the shortener. | QR Code Phishing: Scan to Compromise (Beginner, Free). Also Poster with a Poisoned QR: Event Lures and a Sign-in Relay Kit (2026 track). |
| Smishing | A parcel text: your parcel is held, pay $1.99 to release it. Or an unpaid toll, on a work phone. | An SMS-origin link, a mobile user-agent, an APK download or a card form. No email headers to read. | Check domain age and the redirect chain. Look for an APK behind the link. | FluBot: The Parcel-Delivery Text That Spreads Itself (Beginner, Free). Also The Toll You Never Owed: A Text on a Work Phone (2026 track). |
| Credential harvesting | A login page identical to the corporate one. | A POST to a domain that is not yours, carrying a username and a password. No MFA challenge from your identity provider, because the page is not yours. | Pull the proxy logs and the URL category. Compare the page's tenant branding. | Credential Harvesting: The Lookalike Login (Beginner, Free). Also Who Else Clicked? One Fake Portal, Two Hundred Inboxes (2026 track). |
Stage 2: Access
| Scheme | Victim sees | SOC sees | First move | Practice |
|---|---|---|---|---|
| MFA fatigue | A phone that will not stop buzzing with approval requests. | A burst of denied or ignored pushes, then one approval, often followed by a sign-in from a new device or location. | Count prompts per user in the identity log. Check the device fingerprint of the sign-in that followed the approval. | MFA Fatigue: The Notification Flood (Beginner, Free). |
| AiTM | A normal login where nothing seemed to happen. | A successful sign-in, then the same session cookie used from a new ASN. A mailbox rule created soon after. | Compare session and device across the two sign-ins. Look for token replay, a stolen token used again from another place, rather than a failed password. | EvilProxy AiTM: Indeed Redirect to M365 Cookie Theft (Intermediate, Free). Also Entra ID Device Code Phishing: Token Theft (Intermediate, Free). |
Stage 3: Money
| Scheme | Victim sees | SOC sees | First move | Practice |
|---|---|---|---|---|
| Vishing / callback | A very helpful agent who cancelled a subscription. | A remote-support tool launched minutes after a call the user reports, then a scheduled task or a new service in the same window. | Inventory remote-management tools and their parent process. Line them up against the call timeline. | BazarCall: Callback Phishing to Remote Access (Intermediate, Pro). Also Call This Number: The Debt-Relief Email With Nothing to Click and Support Called on Chat: From a Remote Help Tool to the Certificate Authority (2026 track). |
| BEC (mailbox takeover) | A payment request from the boss, sent from her real address. | A new inbox rule, external forwarding, an OAuth consent (an app the user allowed to read the mailbox), a sign-in from a country the user has never visited. | Audit inbox rules and OAuth grants for the account. Read the sent items, not just the inbox. | Finance Mailbox Takeover at MegaCorp Logistics (Intermediate, Free). |
| Invoice fraud | A vendor who changed banks, on the same invoice, with the same signature. | A new beneficiary on an existing vendor, and a sender domain one character away from the real one. | Diff the sender domain against the vendor master, your approved supplier list with its bank details. Call the number already on file. | Bank Details Changed: The Vendor Email and the Voice That Confirmed It (2026 track). Walkthrough: the invoice that arrived on a Thursday. |
Every scheme links to at least one operation. Those marked 2026 track belong to the new track. If you are new to this, start with a Beginner one: the QR operation covers the whole pattern in one investigation.
Types by how the lure is built
Most lists of phishing types start here. These four names describe who is targeted and how the message is made, and any of them can end as one of the eight schemes above.
| Type | Victim sees | SOC sees | First move |
|---|---|---|---|
| Email phishing | A message from a known brand with a link or an attachment, sent to many people at once. | The same sender, subject or URL across many mailboxes in the message trace. | Scope it: search the trace for the sender, subject and URL, then list who clicked. |
| Spear phishing | A message that names them, their project or their manager. | One recipient or a handful, a sender the tenant has not received mail from before, no campaign pattern. | Check whether that sender has written before, and what the recipient has access to. |
| Whaling | An executive gets a legal notice, a board matter or an urgent approval. | A single high-privilege recipient, and a lookalike domain or a compromised peer account. Often a request, not a link. | Treat it as BEC: check the domain, then the sign-ins and rules on the executive's mailbox and the assistant's. |
| Clone phishing | A message they already received, sent again "with the updated link". | Two messages with the same subject and body, and a different sending server, URL or attachment hash. | Diff the two messages: headers, URL and hash. |
Three more names appear in most lists. Calendar phishing arrives as an invitation file and can sit in the calendar after the email is removed; An Invite Nobody Sent is that case. Search-engine phishing puts a fake login behind an ad or a poisoned result, so the first trace is in proxy and DNS logs, not in mail; Ad, Fake Login, Billing Admin starts there. Angler phishing uses a fake support account on social media and stays outside SOC telemetry until the stolen credentials are used.
The general definition is in the glossary entry for phishing, and social engineering covers the pretexts behind all of them.
The three stages of every fraud
Almost every financially motivated attack runs through the same three stages. The names change. The order does not.
Stage 1: The lure
Something arrives: an email, a text, a QR code on a parking meter, a calendar invite nobody sent, a phone call. The goal is not to hack anything. It is to get one person to click, scan, call or approve.
Owner: Email security
Stage 2: The access
The lure produces a credential, a session, a mailbox rule or a remote-access tool. This is where the SOC has the most telemetry: sign-ins, device IDs and mailbox audit events.
Owner: Identity
Stage 3: The money
The access becomes a payment, a change of bank details, a gift-card run or a ransom. It is the only stage with a hard deadline, and the only one where a phone call can still stop it.
Owner: Finance
Each stage has a different owner. Most SOC teams are organized around the first two and surprised by the third.
The cheat sheet files each scheme under the stage where it pays off, not where its evidence sits. BEC and vishing are filed under money, but the inbox rule, the forwarding and the remote tool they leave behind are stage 2 artifacts.
Three schemes that get misread
Quishing: the page is the payload, not the code
The QR code is not the payload. The page it opens is, and that changes where you look.
An email gateway rewrites the URLs it finds in a message body. A QR code inside a PDF or an image carries no URL to rewrite, and not every gateway decodes images. The user scans it with a phone that is not on the corporate network, and the destination never touches your proxy. The artifacts you have are a PDF and a phone.
The first move is mechanical: extract the code, decode it in a sandbox, follow the redirects to the final domain and check when it was registered. Shorteners are noise. The final host is the evidence.
The physical version needs no email at all. A sticker over a parking-meter code is enough. The only control is in the user's hand: read the domain in the preview before opening it. The track has an operation for that case, Poster with a Poisoned QR, which runs from an event poster to the sign-in relay kit behind it.
AiTM: continuity is the signal
In an adversary-in-the-middle attack the password is correct and the MFA challenge is satisfied, because the user authenticated through the attacker's proxy to the real service.
So the alert is not a bad password. It is continuity: a session that appears from two networks, a token used from a new ASN, an MFA method added from a session that was already authenticated, a mailbox rule created soon after a sign-in the user calls normal.
If you only look for failed logins you will find nothing. The sign-in succeeded. What failed is the assumption that a successful sign-in means the person at the keyboard is who you think.
Device code phishing teaches the same lesson without a proxy. The user types a code into Microsoft's real sign-in page, the attacker receives a valid token, and nothing in the sign-in looks wrong. Entra ID Device Code Phishing: Token Theft follows that case through the Entra sign-in logs, the Graph audit events and the email trail.
Invoice fraud: a procedure stops it, not a tool
By the time the invoice arrives, the phishing may have happened weeks earlier, or there may have been none. Two routes lead here. In one, a compromised vendor mailbox sends a real message in a real thread with a new account number. In the other, an attacker adds a lookalike domain to a real thread and never touches a mailbox.
The control is procedural: a call to a number you already had, not the number in the email. The technical signals support the call: a new beneficiary on a vendor you have paid for years, or a sender domain one character from the real one.
BEC through a taken-over mailbox follows the same pattern. The request comes from a real mailbox and the only anomaly is a rule that hides the replies. Both frauds survive a reading. Neither survives a phone call. The track's Bank Details Changed operation is built on that case: a vendor email, and a voice that confirmed it.
The rule that hides the fraud: one query
In BEC and invoice fraud the attacker needs the victim not to see the replies. A rule that forwards or hides payment mail is one of the clearest artifacts a compromised mailbox leaves, and the FBI warned about auto-forwarding rules used in BEC in a 2020 notice.
This KQL sketch lists inbox rules whose conditions contain payment words. It assumes Exchange audit events reach Microsoft Sentinel in the OfficeActivity table. Adapt the table and field names to your SIEM.
// Sketch: inbox rules that mention payment words. Adapt to your environment.
OfficeActivity
| where OfficeWorkload == "Exchange"
| where Operation in ("New-InboxRule", "Set-InboxRule")
| where Parameters has_any ("invoice", "payment", "wire", "bank")
| project TimeGenerated, UserId, ClientIP, Operation, Parameters
| sort by TimeGenerated descRead Parameters for the rule's conditions and its destination. A rule that moves payment mail to RSS Subscriptions, to Archive or to an address outside the company is the one to escalate. The keyword list comes from published BEC cases. It is a starting point, not a complete list.
To write KQL against the evidence of a case instead of your own tenant, use the Query tab of an investigation. Bank Details Changed and Poster with a Poisoned QR both have one, and Introducing the SOCSimulator Query Tab: Practice KQL in Your Browser explains how it works.
The logs to pull, by stage
| Stage | Pull first | Why |
|---|---|---|
| Lure | Message trace and headers, URL detonation, the raw attachment | Establishes what actually arrived, not what was forwarded |
| Access | Identity sign-in logs, MFA events, device and session IDs, proxy or CASB logs, inbox rules, forwarding and OAuth consents | Shows whether the credential or the session left, and what the attacker set up to stay |
| Money | Payment approvals, beneficiary changes, vendor master | The only stage where a phone call still helps |
If you are triaging a live report, this order doubles as a checklist: lure, access, money. Most investigations stop after the first row, which is why the last one comes as a surprise.
The first ten minutes
When a suspected fraud report lands, the order of operations matters more than the tooling. This is the short version. The phishing email investigation playbook, the BEC investigation playbook and the MFA fatigue investigation playbook have the full steps, and phishing email analysis walks through the headers.
- Preserve the message. Export the EML or the SMS screenshot before anyone replies or forwards. The headers are evidence and they are easy to lose.
- Identify the artifact. Is it a link, a QR code, an attachment, a phone number or a mailbox rule? The artifact decides which scheme in the cheat sheet applies.
- Check authentication. SPF, DKIM and DMARC results tell you whether the sender domain is what it claims to be. They do not tell you whether the sender is trustworthy, and a lookalike domain the attacker owns will pass all three. A message that appears to come from your own domain is worth practising: Automated Transcript: A Voicemail From Your Own Domain.
- Resolve the destination. Detonate the URL or decode the QR in a sandbox. Follow every redirect to the final host and check the registration date.
- Check the identity logs for the recipient. Look for the sign-in that happened after the click, not just the click. Session ID, device and ASN are the fields that matter.
- Scope it. Search the message trace and the click logs for the same sender, subject or URL in other mailboxes. If the message came from a colleague, start from that account as patient zero. Who Else Clicked? One Fake Portal, Two Hundred Inboxes and Patient Zero: The Phish That Came From a Colleague both practise those questions.
- Contain before you conclude. If a session may have been stolen, revoking tokens is more urgent than resetting the password. A password reset does not always end an AiTM intrusion.
- Write the verdict. One paragraph: what arrived, what it did, what it accessed and what you contained. The next analyst will read that paragraph, not your notes.
If two schemes fit, start with the one that has a session or a payment attached. A credential can wait an hour. A payment cannot. If you escalate at the end of a shift, the scheme and the first move you completed are more useful than a paragraph of narrative.
Ten minutes is enough to know which of the eight schemes you have. It is not always enough to close the ticket, but it is enough to stop the wrong one.
Five pairs that get confused
Most misclassification happens between two schemes that look alike. The glossary has a page for each term: Quishing, Smishing, Vishing, Business Email Compromise and Invoice Fraud.
Quishing and smishing. Both arrive on a phone. The artifact is different: an image that has to be decoded, or a link that can be read. If there is no readable URL anywhere in the message, it is quishing.
Credential harvesting and AiTM. Both involve a fake login. In harvesting, the password leaves and the real service never sees a sign-in. In AiTM, the real service sees a valid sign-in and the session leaves. The identity log tells you which one happened.
MFA fatigue and AiTM. Both end with a valid session. In fatigue, the user approves a prompt. In AiTM, the user completes a real challenge through a proxy. If the user says they approved nothing, look for the proxy.
BEC and invoice fraud. BEC is how the request reaches the victim: a business mailbox the attacker controls or imitates. Invoice fraud is what the request asks for: a payment sent to a changed account. One incident can be both. A hidden inbox rule and an odd sign-in point to a taken-over mailbox. A domain one character off and a clean mailbox point to imitation. In both cases the domain diff and the call to the vendor decide.
Vishing and callback phishing. Both involve a phone call. In vishing, the attacker calls. In callback phishing, the email asks the victim to call. The direction of the call changes the first artifact you pull. Call This Number: The Debt-Relief Email With Nothing to Click is the callback case. Passkey Help-Desk Vishing is the case where a caller posing as IT walks an employee through an MFA registration.
Map the schemes to ATT&CK
ATT&CK describes the intrusion, not the payment, so a diverted payment has no technique of its own. Our MITRE ATT&CK library has detection pages for several of them, including T1566 Phishing and T1621 MFA request generation.
| Scheme | Technique | Note |
|---|---|---|
| Quishing | T1660 Phishing, T1598.003 Spearphishing Link | ATT&CK names quishing under both; there is no QR-only technique |
| Smishing | T1660 Phishing | The mobile technique |
| Credential harvesting | T1598.003 Phishing for Information: Spearphishing Link | |
| MFA fatigue | T1621 Multi-Factor Authentication Request Generation | |
| AiTM | T1557 Adversary-in-the-Middle, T1539 Steal Web Session Cookie | |
| Vishing / callback | T1566.004 Spearphishing Voice, T1219 Remote Access Tools | The second for the tool installed after the call |
| BEC (mailbox takeover) | T1114.003 Email Forwarding Rule, T1564.008 Email Hiding Rules, T1078.004 Cloud Accounts | |
| BEC (lookalike domain) | T1684.001 Social Engineering: Impersonation | Added in April 2026 |
| Invoice fraud | None for the diverted payment | The way in maps to the two BEC rows above |
A worked example: the invoice that arrived on a Thursday
This is a simulated scenario, not a real case. The schemes are easier to remember with a timeline attached.
- Thursday, 16:40. A finance analyst forwards an email to the SOC queue from a supplier the company has paid for four years. The invoice is attached. The bank details have changed, and the note explains that the old account is closed. The display name is correct and the thread has six messages going back to the previous quarter.
- 16:52. The SOC analyst checks the sender domain against the vendor master. It is one letter off the real one. That is the first artifact, and it points at two schemes at once, BEC and invoice fraud.
- 17:05. The audit of the finance analyst's mailbox shows a rule created eleven days earlier that moves anything containing "invoice" or "payment" to an archive folder and forwards a copy to an external address. The sign-in log shows a login from a new ASN on the same day the rule appeared, with MFA satisfied. That is the access stage, and it happened more than a week before the invoice arrived. The attacker had been reading the thread in that mailbox and replayed it from a lookalike domain.
- 17:20. The payment was scheduled for Friday at 09:00. The finance team calls the number already on file, not the one in the email. The vendor's controller confirms that the bank details have not changed. The payment is cancelled before it leaves.
- Friday, 09:00. Nothing happens, which is the point.
Two schemes applied, and both were true. The compromise was a mailbox, so BEC explained the rule, the forwarding and the login from a new network. The attempt was a diverted payment, so invoice fraud explained the beneficiary change and the one-letter domain. The first move that identified the scheme was the domain diff. The move that stopped the money was the phone call.
The rule had existed for eleven days before anyone looked. The story ended well only because the payment run had a deadline the team could still meet. You can work a case of this shape in Bank Details Changed: The Vendor Email and the Voice That Confirmed It.
What this cheat sheet does not cover
- Ransomware and extortion. They often start with a phishing email, but the fraud is a different business model and the response is a different playbook.
- Consumer scams. Investment fraud and tech-support scams are two of the largest categories by loss, but they rarely land in a SOC queue.
- Pasted commands, ads and fake recruiters. They start as fraud and end as malware or cloud abuse. The track covers them in The Assistant That Said Paste This, Ad, Fake Login, Billing Admin and The Interview That Compiled. Prompt in the Footer covers a phish aimed at an AI assistant.
- Deepfake video and voice. The pattern is close to vishing. The detection is not. It deserves its own piece.
Knowing what is out of scope is part of using a cheat sheet. This one covers the schemes that reach an analyst's queue.
Pick your first operation
You do not need a whole track to start. Pick the scheme that sounds most like your week.
- If your users scan things: QR Code Phishing: Scan to Compromise (Beginner, Free).
- If your users approve push prompts: MFA Fatigue: The Notification Flood (Beginner, Free).
- If finance lives in one mailbox: Finance Mailbox Takeover at MegaCorp Logistics (Intermediate, Free).
- If the report is just a callback number: Call This Number: The Debt-Relief Email With Nothing to Click (2026 track).
- If a vendor just changed bank details: Bank Details Changed: The Vendor Email and the Voice That Confirmed It (2026 track).
Not every phish has a link to click
2026 Phishing: Beyond the Inbox starts from a callback number, a text or a QR poster: 17 operations, most built on attacks reported between August and November 2026.
International Fraud Awareness Week 2026 runs November 15 to 21. This piece is part of it, with new glossary terms and the new track.
Train on real alerts, with zero consequences
Practice triage on realistic alert volume in a live SOC console. Free.
Frequently asked questions
What are the main types of phishing attacks?
By channel: email phishing, smishing by text, vishing by phone call and quishing through a QR code. By what the attacker takes: credential harvesting collects a password, adversary-in-the-middle phishing steals the session after MFA, and MFA fatigue pressures the user into approving a prompt. Business email compromise and invoice fraud are where many of them end, with a payment request from a mailbox the attacker controls or imitates.
What is the most common fraud in 2025?
By reported losses, investment fraud led the IC3's 2025 report at $8.6 billion, followed by business email compromise at $3.05 billion and tech-support scams at $2.1 billion. By number of complaints, phishing and spoofing came first, with 191,561.
What is the difference between BEC and invoice fraud?
BEC is fraud through a business email account the attacker controls or imitates. Invoice fraud is the diversion of a payment by changing the bank details on a bill. They overlap, because most invoice fraud runs through a compromised or imitated mailbox.
What is the difference between quishing and smishing?
Both reach the victim on a phone. Smishing is a text message with a link you can read. Quishing hides the link in a QR code, usually inside an image or a PDF, so a text-only filter never sees it.
How do you detect an AiTM attack?
Look for continuity, not failure. A successful sign-in is followed by the same session or token from a new ASN or device, an MFA method is added from an already authenticated session, or a mailbox rule appears soon after a sign-in the user calls normal. Revoke the sessions, because a password reset alone may not end it.
What should I do first if I suspect invoice fraud?
Call the vendor on a number you already had, not the one in the email, before the payment runs. Then preserve the message, diff the sender domain against the vendor master and check the mailbox for new rules. If money has already left, tell the bank at once; in the US, also file a complaint at ic3.gov.
Where should I start?
With the operation that matches something you have already seen. QR Code Phishing: Scan to Compromise is a Beginner operation and shows the whole pattern, from the lure and a stolen session to a hidden forwarding rule, in a single investigation. If your queue is mostly reported emails, the 2026 Phishing: Beyond the Inbox track opens with an easy one, After the Breach: The Compensation Email.
Is there a track to practice fraud and phishing investigations?
Yes. 2026 Phishing: Beyond the Inbox has 17 operations, most built on attacks reported between August and November 2026. Most never touch a plain email link, so you work them from the logs, not from the inbox.
Nicole researches and structures SOCSimulator's blog guides and product announcements, from SOC playbook templates and the differences between EDR, antivirus and XDR to the launch of the KQL Query tab. Nicole also looks after how each post is found in search, so the answer an analyst needs is the one they actually find.


