Planning your move from KnowBe4? Start here

Threat Intelligence

Stage 5, Compromise: When the Damage Is an Authorized Action

Compromise is where social engineering damage lands as an authorized action: a real login, consent, ticket, or payment. See why it’s hard to spot and where controls work.

Stage 5 Compromise hero image

The record of a successful social engineering attack rarely shows anything breaking. It shows a valid session, an approved prompt, a closed support ticket, or a payment that someone with authority released, because producing those events is what the earlier stages were built to do. This piece walks through Compromise, the fifth stage of the social engineering attack chain: what the adversary actually comes away with, why the result looks like ordinary business and so rarely ends the operation, and where a control can still say no.

The incident record will say that an account was compromised, a payment was redirected, or a credential was reset. The system logs will say that an authorized user took an authorized action. Both statements are accurate, and the distance between them is the stage this piece is about.

The earlier stages were built to produce one result: a person, or a process, doing what it was designed to do, at a moment the adversary chose. That result is Compromise, the fifth stage of the chain, and it is where the damage is done by an action your systems were built to allow.

The first piece in this series argued that social engineering is a chain, not an event. The second followed the money. The four that came after walked the buildout, the moment it went live, the crossing that reached a person, and the exchange that person was drawn into. This piece follows that exchange to its result.

What Compromise actually is

In the five-stage model, Compromise is when the adversary’s objective lands. The model defines the stage by what the operation obtains, not by how it obtains it, which is why a single definition covers a stolen session, a redirected payment, a reset credential, and an approval that opens an account.

For most of the history of the field, compromise meant a password. Today the password is often the smaller prize. The same exchange that could have produced a credential can produce a live session, standing access, or a payment instead, and each of those can be worth more to the operation than the password ever was.

What attackers actually walk away with

The outcomes fall into a handful of recognizable kinds, and they frequently involve no malware at all.

Sometimes the prize is a live session. In an adversary-in-the-middle attack, the target types a real password into a relay, approves a real MFA prompt, and lands on the page they expected. The relay keeps the session token that the real service issues in response. The authentication was genuine, so the login reads as a success, and the adversary replays the token from their own device. A password reset alone may not end that access, which is why first-hour containment begins with revoking the sessions and refresh tokens and only then resets the credential.

Sometimes it is standing access. The employee signs in at the real login page, completes the real MFA prompt, and then sees a consent screen from an application wearing a trusted vendor’s logo. Approving it gives the application programmatic access to mail and files with no password involved, and a password reset does not necessarily remove it. The sign-in itself was not attacked, because it was genuine.

Sometimes it is a decision inside a process. A threat actor does not need a credential if a support workflow will issue one. A password reset, a factor reset, or a device re-registration approved on the strength of a good pretext is the whole compromise, and the system records it as a resolved support ticket.

Sometimes it is a payment. The invoice, the amount, and the email thread are all real, and what changes is the receiving account. The approval that releases the money is genuine as well, and a second approver who reviews the same unverified request adds a second signature without adding a second check.

What these outcomes share is that each one is an authorized event. The session came from a real login, the access from a real consent, the reset from a real ticket, and the payment from a real approval. The adversary did not defeat a control. The operation arranged for a control to say yes.

Why Compromise is hard to see

The result looks like the system working. A successful MFA approval is treated as proof of safety rather than as the start of an investigation; a closed ticket reads as service delivered, and an approved payment reads as process followed. Detection is tuned to catch something that failed, and in these outcomes, little failed.

The evidence also lands in different places. The identity team sees a login, the help desk sees a ticket, and finance sees a payment, and rarely does any one of them hold the sequence that connects the three. In the reconstructions I have worked, the compromise itself is seldom what gets noticed first. The first sign is more often a consequence somewhere else: a vendor asking about a payment, or a customer replying to a message the account owner never sent.

Then there is time. When nothing alarms, nothing starts a clock, and a compromise that goes unflagged can run for months. Mailbox rules that hide replies and move messages away from their owner keep the account usable and quiet while the operation uses it.

What happens after Compromise

Compromise is the last stage in the model. It is seldom the last move in the operation. What the adversary holds at this point is more than the outcome the exchange was built to produce. It is a trusted position inside your organization and among the people who deal with it, and that position is often a better staging ground than one the operation could build from scratch.

Read in terms of the chain itself, the position supplies most of what the next operation needs. A compromised mailbox sends from a real address, inside real threads, with authentication signals that can all pass, which makes it better Launch infrastructure than a domain an adversary can register. A harvested directory answers the target-selection questions that Setup had to research. A thread history supplies the pretext that Contact and Engagement would otherwise have to invent. The output of this stage is the input to the next chain, and the next target can be a customer, a vendor, or a colleague of the person who was just compromised.

This is where the series draws a line on purpose. Frameworks such as MITRE ATT&CK (opens in new tab) already catalog what an adversary does once it holds a foothold inside the environment: persistence, lateral movement, exfiltration, and impact. The attack chain maps the other side of that boundary: how the foothold was obtained from a person, and what the operation does with the trust it earned. Treating Compromise as the point where the two meet keeps both maps intact.

What defenders get wrong

The common mistake at this stage is treating Compromise as an event to remediate instead of the pivot into a wider operation. A reset credential and a closed ticket feel like the end. Resets and containment are necessary, and the order matters. An incident scoped only to the account that was reset still leaves the important questions open: what the account did between the compromise and the discovery, who it wrote to, and where the operation that produced it went next.

The second mistake is reading a clean authentication as a clean outcome. A successful login with a passed MFA check answers whether the person who authenticated is allowed in. It does not answer whether the person who asked them to act is who they claim to be, and that is the question this stage turns on. Authentication can confirm that an employee may initiate a payment, and it cannot confirm who instructed them to.

Where controls still work

The adversary needs a control to say yes, so the controls that matter most sit at the actions the operation is built to obtain.

Put verification on the action, not on the person. A payment-destination change, a factor reset, a new application consent, and a new mailbox forwarding rule are the moments the operation needs a yes. Each can carry a check that does not depend on anyone’s judgment under pressure: a callback to a number the organization already held, a second reviewer who looks at independent evidence instead of the same request, an administrator review for new application consents. The check belongs to the process, so the person at the keyboard is not asked to improvise it.

Watch what follows the approval. Detection at this stage is sequence-based: the question is not whether the login succeeded but what happened after it. A session created from a new context moments after a genuine approval, a new rule that moves replies out of the owner’s view, and a first consent to an unfamiliar application are all facts about what followed the yes, and each one can be logged and correlated.

Contain by what the access could reach. Revoke the sessions and refresh tokens, remove the consents, rules, and delegated access, and only then reset the credential, because a reset made first can tip off an adversary who still holds a live session.

Trace the compromise back to its campaign, then forward to everyone it reached. The compromised account was one target of an operation that staged its infrastructure during Setup, and correlating the compromise with that campaign can turn one incident into a view of who else received the same lure, which vendors and customers the account wrote to, and what infrastructure can be taken down before the next hop.

Why the authorized action is the control point

The instinct this stage punishes is waiting for something to break. The login is real, the consent is real, the ticket is real, and the payment is real, so a program organized around catching a failure has very little to catch. The posture that works accepts that a person at the keyboard will sometimes be persuaded, and it builds the decision into the process instead: verification attached to the action, visibility into what follows it, and a fast path from one compromise back to its campaign. The adversary is counting on those approvals staying private judgment calls. Make them a process decision, and the operation has to get its next yes past a control that asks for proof.

What leaders should take away

  • A successful attack can look like normal operations. A valid session, an approved prompt, a closed ticket, and an authorized payment are the outcomes the operation is built to produce, and they rarely generate a failure to alert on. A program that waits for something to break has little to see at this stage.
  • The control point is the action, not the login. Payment-destination changes, factor resets, application consents, and mailbox rules are where the operation needs a yes. Verification built into those actions, and run outside the conversation that requested them, protects the organization without depending on anyone’s judgment under pressure.
  • Compromise is more often a beginning than an end. A stolen session, mailbox, or directory is staging for the next operation, and the next target can be a customer, a vendor, or a colleague. Programs that trace a compromise back to its campaign, and forward to everyone it reached, limit the next launch before it starts.

The next piece in the series steps back from the individual stages to ask why defenders keep missing the chain itself: how an organization built to see fragments ends up facing an operation that spans all five.

Learn how Doppel can protect your brand from social engineering attacks

Join hundreds of companies already using our platform to protect their brand and people from social engineering attacks.