Skip to main content
The Onlooker SeriesChapter 2

Chapter 2: The Authority Gap

How iVALT places a verified human between stolen identity and consequential action

By Chuck Mackey, iVALT6 min read

In brief

Chapter 1 named the Onlookers, the ordinary people criminal enterprises borrow. This chapter names the gap that makes the borrowing work. Identity systems prove that an accepted credential was presented; they do not prove that the verified human who holds the authority approved the specific action. That distance is the authority gap. This chapter walks the seven decision gates where a credential alone should never be enough, shows how one verified human breaks the workforce, communications, financial, and AI chains, and states plainly what iVALT does not eliminate.

The request arrived at the worst possible moment.

A senior employee was traveling. The help desk was busy. A caller needed a password reset immediately and possessed enough personal information to survive the organization’s standard verification questions.

The employee number was correct. The manager’s name was correct. The caller knew the last project discussed in email. The request originated from a plausible location.

Under a traditional identity system, the evidence could be sufficient.

Then the process stopped.

Not because the password was wrong. Not because the caller’s story had been disproved. Not because an anomaly engine had calculated an unacceptable probability.

The action stopped because the organization required something the caller could not produce: proof that the real, authorized human was present and approving the action.

That is the control boundary iVALT is designed to create.

Identity is not authority

Most identity systems answer a narrow question: does the requesting party possess an accepted credential? The Onlooker threat exposes the weakness in that model. A credential can be correct while the person using it is wrong. A session can be valid while the human authority behind it has changed. A trusted employee can execute an action after being deceived. An AI agent can act under delegated permissions without proving that an authorized human approved the particular action.

Chapter 1 showed an enterprise that borrows trust. The borrowing works because systems verify artifacts: something presented, something recognized, a workflow state that looks complete. Authority is a different kind of fact. It belongs to a person. It attaches to a specific action. It does not transfer with a stolen login, and it does not survive being borrowed. A system that cannot tell the difference will keep approving the wrong human, confidently and at scale.

The gap is easy to miss because the system reports success. The log shows an authenticated user, an approved action, a completed workflow. Every entry is accurate, and the conclusion drawn from it is wrong, because nothing in the record says the human who held the authority was ever there. That distance, between the credential accepted and the approval never given, is the authority gap.

iVALT is positioned as a human-authority layer placed before consequential actions. Instead of relying on a credential alone, it is designed to seek assurance that an authorized human is present and responsible before the protected action executes.

Did the verified human who possesses the authority approve this specific action?

The gates

Consider the attacker who has stolen enough of an employee’s digital life to pass as that employee to most systems. The evidence is assembled. The first door opens. A human-authority control changes what that evidence is worth: possession may get the attacker into the account, but it does not authorize what happens next.

What happens next is a short list of moments where a credential, on its own, should never be enough. Password resets. New authorization methods. Device enrollment. Privilege elevation. Payment instruction changes. Large transaction approvals. Privileged AI agent actions. Seven gates. Each one is a place where stolen identity converts into consequence.

  1. 1Password resets
  2. 2New authorization methods
  3. 3Device enrollment
  4. 4Privilege elevation
  5. 5Payment instruction changes
  6. 6Large transaction approvals
  7. 7Privileged AI agent actions

Everything else can move at the speed of the workflow. These seven are where the workflow waits for a human. Notice what the list leaves out: reading email, opening a document, joining a meeting, the ordinary work of the day. The gates are not a tax on everything. They are a lock on the few actions the attacker actually came for.

The stolen identity may still open a door. But the attacker encounters another door before the action that matters.

Breaking the workforce chain

The workforce Onlooker is dangerous precisely because nothing appears to be wrong. The employee is fully authenticated and still manipulated. A help desk analyst receives a convincing request to disable multifactor authentication, logs in legitimately, follows the approved procedure, and completes it. From the system’s perspective, the process worked exactly as designed.

A human-authority control changes the analyst’s job. The person whose account is being changed can be required to prove presence directly, through a separate trusted channel, before the help desk completes the request. The analyst no longer has to decide identity based on knowledge, voice, urgency, or plausibility.

Training asks a person to resist deception. Compensating control reduces the value of the deception.

Breaking the communications chain

A message from a compromised mailbox can be authentic in every technical sense. It came from the real account. It sits inside the correct conversation. It contains accurate context. Email authentication cannot resolve that problem, because the message may have passed through the authorized system on its way to the inbox.

The proof has to move outside the channel. A finance employee who receives changed supplier instructions does not have to trust the email merely because it came from the supplier’s real domain. The workflow can require a verified human associated with the supplier to approve the change before it takes effect. The attacker can control the mailbox and still fail to control the human. The borrowed credibility of Chapter 1 stops working the moment approval has to come from somewhere the attacker does not control.

Breaking the financial chain

Business email compromise usually succeeds before any money begins moving. The decisive moment is earlier and quieter: the change to a supplier record, the new payee, the exception that bypasses normal procedure. By the time a transfer looks suspicious, the authority behind it has already been borrowed.

So the control does not try to inspect every financial event equally. It defines protected decision gates, starting with supplier banking changes and high-value approvals, where credentials and workflow status are insufficient without direct human authority. Fewer checks, placed exactly where the consequence lives.

Breaking the AI chain

The next generation of the Onlooker problem may begin with no stolen password at all. An AI agent operating under valid delegated permissions can generate communications, trigger workflows, access data, initiate transactions, and interact with other agents. If that agent inherits authority from a user, a role, or a token, the organization may be unable to demonstrate that a real person approved the specific consequential action. The human becomes an Onlooker to an automated process operating in the human’s name.

The answer is not a person standing in front of every AI operation. It is a verified human sponsor at the boundary where automation becomes consequence, required before the agent performs sensitive work. The sponsor is not there to review the agent’s reasoning. The sponsor answers one question: is a human who holds the authority approving this action, now?

What iVALT does not eliminate

No single identity control eliminates the Onlooker problem. Human verification does not patch vulnerable routers. It does not remove malware from endpoints. It does not prevent every employee from being deceived. It does not guarantee that an authorized person is making a wise decision. A verified human can still act negligently, maliciously, or under coercion.

iVALT is best understood as a compensating control for the authority gap. It reduces reliance on credentials, sessions, device possession, and social engineering resistance at the moments where the consequences are greatest.

The cybercriminal may steal the user’s name, password, token, session, mailbox, or device context. But before the criminal converts stolen identity into a consequential action, the organization can require the verified participation of the authorized human.

The attacker may possess an identity. The remaining question is whether the attacker can prove the human authority required to act.

Chapter 3, The Document That Refused to Become an Onlooker, follows: the same principle applied to the file itself, as DocuID makes protection travel with the document.

For Security & Risk Leaders

Find your seven gates.

Map the consequential actions in your organization, and where verified human authority belongs, with the iVALT team.

Request an Executive Briefing

Frequently Asked Questions

What is the authority gap?

The authority gap is the distance between what an identity system proves and what an organization actually needs to know. The system proves an accepted credential was presented. The organization needs to know that the verified human who holds the authority approved this specific action. Stolen credentials, valid sessions, deceived employees, and AI agents acting on delegated permissions all fall into that gap.

How is identity different from authority?

Identity answers who a credential claims is acting. Authority answers whether the person entitled to act is the person acting, for this specific action. A credential can be correct while the person using it is wrong, and a session can be valid while the human authority behind it has changed. Identity is a claim about an account. Authority is a fact about a human, and it does not transfer when a credential is stolen.

Which actions should require verified human approval?

The seven decision gates where a credential alone should never be enough: password resets; new authorization methods; device enrollment; privilege elevation; payment instruction changes; large transaction approvals; and privileged AI agent actions. These are the moments where stolen identity converts into consequence, so these are the moments where a verified human has to stand.

What does iVALT not eliminate?

No single identity control eliminates the Onlooker problem. Human verification does not patch vulnerable routers, remove malware from endpoints, or prevent every employee from being deceived, and it does not guarantee that an authorized person is making a wise decision. A verified human can still act negligently, maliciously, or under coercion. iVALT is a compensating control for the authority gap, placed at the moments where the consequences are greatest.