How HHS Is Using AI to Strengthen Financial Oversight

Presented by Appian

In his conversation with Francis Rose on Beyond the AI Pilot: Turning Investment into Mission Outcomes, presented by Appian, Chiarello explains how the initiative connects existing audit information to a more coordinated approach to oversight.

Chiarello says the initiative grew from examining recurring weaknesses in oversight of grant-funded programs. Reviewing audit findings led him to ask why problems identified in earlier years had not produced a clearer response. The issue, as he describes it, involved a federated process in which information moved among divisions and their counterparts without necessarily providing a coordinated view of unresolved deficiencies.

In that environment, a subsequent audit could become the next opportunity to learn whether a weakness had been addressed. That left leaders looking for a better way to understand repeated findings and act on them.

Large language models provide a way to examine substantial amounts of audit material and identify patterns over time. Chiarello emphasizes that the underlying information was available already. The change is the ability to read, compare, and interpret it more efficiently.

He contrasts that capability with earlier efforts that depended on paper records. Information could exist in abundance while remaining difficult to use for broader analysis. With newer tools, reviewers can examine similarities in language, the persistence of weaknesses, and whether reporting indicates that a problem is improving.

The resulting analysis can help officials identify situations that deserve closer attention. Chiarello is careful to separate an indicator of possible fraud from evidence establishing that fraud occurred. Patterns can support a decision to investigate further; they do not eliminate the need for human review or an appropriate evidentiary basis.

That distinction matters because oversight decisions affect real services and the people who depend on them. Chiarello describes the need to exercise judgment when considering responses to missing audits or other deficiencies. A grantee that is the only healthcare provider within a large geographic area presents different consequences from a case where withholding funds would not create the same service disruption.

His account connects AI analysis to the practical responsibilities of program management. Finding a risk is one step. Determining what it means, discussing corrective action, and considering the effect on beneficiaries remain essential parts of the work.

Chiarello also sees potential beyond HHS. Other departments face comparable responsibilities for overseeing federal funds, making the approach a possible foundation for wider use. He describes that expansion as an objective still being developed, with the model and implementation instructions not yet complete.

For agencies looking for tangible mission outcomes, the interview offers a useful example: begin with a defined responsibility and an existing body of information. Use AI to improve visibility, then connect that visibility to deliberate human decisions about accountability, corrective action, and the continued delivery of essential public services.

3 Key Takeaways

• AI can help officials analyze existing audit records for recurring weaknesses and emerging risk patterns.
• Risk indicators should guide further review; they do not independently establish fraud.
• HHS is developing the approach with a goal of broader use across government.