The Department of Housing and Urban Development (HUD) is planning to launch an AI tool that reviews how grant recipients spend federal money, with a target launch date of September 30. 

HUD’s acting chief financial officer (CFO), Irving Dennis, said the goal is to eliminate waste, fraud, and abuse, according to a FedScoop report. The tool will pull invoice and contract data for every check a grantee writes to a vendor.

According to the report, HUD sends out roughly $77 billion a year to grantees, but Dennis told the Senate Banking Committee in July that the agency doesn’t currently know how that money is spent once it’s out the door. 

The system is part of the HUD Unified Grants System, known as HUGS. HUD awarded a $500,000 contract to Palantir to help build it, and a pilot program testing the tool on 10 grantees was, in Dennis’s words, “a great success,” according to FedScoop.

Keeping a Human in the Loop

Dennis said in written responses to Senator Elizabeth Warren that “HUD intends for human review and oversight to remain integral to grants-management decisions,” and that the AI components are meant to “support, not replace” staff judgment, according to FedScoop.

A centralized data science team inside HUD’s CFO office will monitor the tool’s output, and any final determination will be made by a person, according to a HUD source.

Quinn Anex-Ries, senior policy analyst with the Center for Democracy & Technology (CDT), told FedScoop the tool appears to meet the criteria for a “high-impact” AI system under Office of Management and Budget (OMB) guidance, which calls for specific risk management practices before launch and for public AI inventories to be kept current. 

As of the report, HUD hadn’t yet updated its inventory to include the tool.

What This Signals for Building AI Oversight the Right Way

For chief data officers (CDOs), HUD’s plan is a live test of a problem every large organization deploying AI at scale eventually faces: written governance commitments only matter if they’re followed consistently once the system is running. 

HUD says humans stay in the loop and that a dedicated team will monitor the tool’s output, but the actual safeguards, and whether HUD’s public AI inventory gets updated before launch, as OMB’s own high-impact guidance requires, are yet to be resolved as the launch date approaches.

That gap between stated policy and demonstrated practice is the same AI governance risk CDOs manage internally: a governance framework that exists on paper but isn’t verifiably enforced offers little real protection, whether the AI system in question flags a grant recipient’s spending or an internal enterprise process. 

HUD’s rollout, and how closely its actual practices match its written commitments, is worth watching as a signal for how seriously other large organizations are treating that same gap.