Catch What Background Checks Miss:How Alight Defends Against Candidate Fraud

Fraud doesn't announce itself. By the time a background check flags something, the organization has already run someone through screening, invested recruiter time, and sometimes started pre-boarding — only to begin the whole cycle over.

For Alight Solutions, a benefits administration company that manages health, wealth, and wellbeing programs for some of the world's largest employers, that risk carries extra weight. Alight's clients trust them with sensitive benefits data for 36 million people. 

At IAMPHENOM 2026, Julie Eagy, Talent Acquisition Operations Manager at Alight, shared how the company is in the early testing phase of Phenom Fraud Detection Agent. Find out what they set out to solve, what the data revealed, and how they're building toward a full deployment.

Watch the full session here, or explore the highlights below!

What Types of Candidate Fraud Are Common in High-Volume Hiring?

Alight's high-volume seasonal hiring is built for speed and scale. "Up to the point of offer, a candidate is not interacting with anyone on our team directly. There's no face-to-face or video interview — it's all done through digital screening, and we make decisions right from that," said Eagy. While that low-touch model works well for those positions, it also leaves room for fraudulent candidates to enter the pipeline.

Eagy described a pattern her team had already spotted manually: the same person applying under two different names. The recordings made it identifiable (if you happened to be looking), but with multiple recruiters independently working through copious applications, it was easy to miss.

Shared resumes were another recurring issue. Different candidates would submit near-identical documents with minor changes across separate requisitions. Without a central system to surface the overlap, those duplicates moved through undetected. Background check vendors, by design, sit too late in the process to catch either problem.

How Does Phenom's Fraud Detection Agent Work?

Phenom Fraud Detection Agent works across the full hiring journey: application, interview and screening, and pre-boarding. 

At the application stage, the agent flags duplicate resumes, repeated submissions under different aliases, and applications from regions that don't match role eligibility. During interview and screening, Fraud Detection Agent integrates with Phenom Interview Intelligence to analyze candidate interactions during live and recorded video interviews, detecting face and voice inconsistencies, script-reading behavior, and mid-interview identity swaps. At pre-boarding, the focus shifts to confirming that the person who completed the hiring process is the same one showing up on day one.

A core design principle is that the agent flags rather than decides. AI fraud detection is probabilistic, so false positives are possible. Every flag is accompanied with the underlying reasoning and timestamped evidence, including the specific recording moment, transcript excerpt, or document comparison, so recruiters can review and make the call themselves.

The agent also tracks what the team calls compound malpractice. If a candidate appears on a do-not-hire list and has also triggered identity flags, those signals are combined into a fuller picture rather than treated as separate incidents.

Related: Meet the Fraud Detection Agent: Building Hiring Confidence Through Intelligent Verification

What Can AI Fraud Detection Reveal During Hiring?

To build confidence in the agent before any broad rollout, Alight started with a structured testing phase. They shared recordings and application data from four closed requisitions with the Phenom team and ran them through the agent, then reviewed the findings themselves to validate what it caught.

The testing surfaced two distinct cases. In the first, the agent was pointed at known fraud Alight had already identified manually — the same individual who had screened under two different names, a pattern multiple recruiters had missed. In the second, 1,526 candidates were analyzed for duplicate applications under different names, with initial flags narrowed down to confirmed identity mismatches after human review.

That review process also drove calibration. Early on, the agent flagged candidates glancing off-screen during a screening. Alight determined that wasn't a reliable signal, and the sensitivity was adjusted accordingly. By the end of the testing phase, the numbers reflected how much ground the agent had covered:

• 175 candidates analyzed across four requisitions during the initial testing phase

• 537 warning signals detected spanning identity, behavioral, and conversational anomalies

• 87% facial match confidence achieved on known impersonation cases

• 134 warnings validated through human review

• 85%+ human-validated confidence rate across flagged fraud detection cases

"Most of what the agent found, especially the high-severity cases, were definitely something that we would stop the process with that candidate. We found it to be pretty reliable," said Eagy.

What AI Hiring Fraud Detection Features Are Coming Next?

The roadmap for Fraud Detection Agent continues to expand.

Automatic removal of flagged applications is on the horizon, reducing the manual effort required to act on detected signals. Any data collected during fraud detection, including recordings, transcripts, and flagged signals, is retained for 12 months by default. Customers can adjust that window based on their own compliance requirements, including GDPR.

There's also interest in a cross-platform blocklist, where a candidate flagged by one Phenom customer could be visible to others. That's technically possible, but the legal and privacy framework for doing it responsibly is still being worked out. Longer term, a natural-language policy engine will let customers set their own rules directly in the platform without a support ticket. Customers will be able to define a signal, pair it with an action, and let the agent handle the rest.

For Alight, the testing phase was just the starting point. As the agent continues to evolve and deployment scales, the goal stays the same — know who you're hiring before they walk in the door.

Don’t miss our webinar, Candidate Fraud Detection in the Age of AI: How to Catch What Traditional Hiring Can't