This post is co-authored by Pega and Celebrus, learn more about the partnership.
Sarah books a flight. Not quite.
She opens her phone, asks her AI assistant to find the best fare to Barcelona for the last week of June, and puts it down. A few minutes later, the ticket is confirmed. No form filled out. No card number entered. Her agent handled it, authenticated, delegated, and done.
For the airline, that transaction looked like any other customer interaction. But the behavioral signals were different. The session pattern was different. The identity was presented differently. The booking was real and the payment cleared, but the experience on the other side of that channel was something entirely new.
Now imagine the same scenario, but it wasn't Sarah's agent.
That shift is already underway. AI agents are booking travel, completing applications, processing payments, and managing subscriptions on behalf of the customers brands work hard to serve. For fraud teams, brand leaders and decisioning executives, that shift creates a question that most existing systems were not designed to answer: who is acting, with what authority, and against what evidence?
Three risks agentic AI creates
The Sarah scenario is benign. But the same infrastructure that makes her agent possible: delegated authentication, cryptographic identity, autonomous action across channels, creates new exposure at exactly the points where brands are least prepared. Three risks stand out.
Fraud. Attackers were already probing across sessions, devices, and channels. Agents make that easier, faster, and cheaper. Scripted activity that used to be obvious now blends into traffic that can look signed, legitimate, and authorized.
Decision opacity. When an action is taken by an agent, “the customer authorized it” is no longer a complete answer. Was the authority delegated correctly? Is the behavior consistent with how this person transacts? Did the decisioning engine explain itself?
Brand Safety. The agentic web is rapidly turning into a question of which agents your brand admits, which it challenges, and how proportionately it intervenes when something looks wrong. Get it wrong in either direction and the cost show up in the wrong place lost legitimate journeys, or fraud absorbed in silence.
Connected customer journeys are safer than disconnected ones
Fraud rarely happens in a single moment. It happens across a journey. The same is true for trust. When behavioral evidence and identity persist across sessions, devices, anonymous interactions and authenticated logins, signals compound. When they are siloed, each tool sees a fragment, and patterns slip through the gaps.
What makes those patterns visible is customer identity resolution: connecting behavioral signals across anonymous and authenticated states into a single, continuous view of the person.
Making those patterns visible requires data that is first-party, real-time, and identity-persistent across all three states a customer moves through: anonymous, known-but-logged-out, and authenticated. The behavioral dimension matters, too. Typing rhythm, navigation patterns, hesitation, inter-click timing: these micro-signals are unique to how real people move, and they look vastly different from scripted or delegated activity. When that evidence is captured at source and held continuously across the journey, the decisioning layer never encounters a stranger.
From real-time behavioral signals to responsible AI decisions
Think back to Sarah’s booking. The airline saw a completed transaction. What it didn’t see was whether that session matched her usual behavior, whether the identity presented was consistent with her history across channels, or whether the action taken was proportionate to what the profile warranted. That is the gap Celebrus and Pega CDH close together.
Celebrus captures real-time behavioral signals and resolves identity continuously across the journey, so that when Sarah’s agent, or someone pretending to be it, arrives mid-journey, the system already knows what normal looks like. Pega Customer Decision Hub then acts on that evidence: not just applying a rule but reading the profile alongside the behavior to determine the right action in that moment: a retention offer, a step-up challenge, or simply letting the journey continue. Celebrus is rich on signals and deliberately neutral on verdicts. The decisioning engine of record applies policy, makes the call, and is designed to support accountability. Keeping them separate matters. Behavioral and identity evidence flows in at speed. Pega’s responsible-AI controls wrap the action that goes back out: transparency, fairness, oversight, and ethical guardrails.
The goal is proportionality: agents operating under valid delegation should move through without unnecessary friction; suspicious sessions, human or otherwise, should be stepped up rather than blocked outright. Legitimate customers, and the agents acting on their behalf, keep them journey. Brands keep both their experience and their compliance posture.
What agentic AI means for identity, fraud strategy, and decisioning going forward
Agentic AI raises the stakes for identity and fraud strategy rather than replacing them. Capture the evidence at source, connect the journey, and let a responsible decisioning layer act on it. That is how brands stay safe and stay open as the next class of users, human and agent alike, turns up at the door.
Sarah’s flight got booked in minutes. The question for every brand is whether, the next time that happens, they’ll know it was really her.