The US government has misspent almost $3 trillion since 2003, leading government auditors to make a fresh call for a federal data analytics centre – a proposal first suggested in 2022.
Total improper payments reached $185.8 billion in 2025 alone, according to a report by the Government Accountability Office (GAO) – up $24 billion on 2024 in the first rise for the statistic since 2021.
Overpayments by federal agencies were the largest problem, accounting for 82% of the misspending, and just five programmes misspent $135.8 billion, with Medicare making $57 billion in improper payments.
The real number for money lost is likely much higher, added GAO, which estimated the government loses up to $521 billion annually to fraud.
The problem “demands urgent action,” said GAO Head Orice W. Brown, “agencies need stronger controls, better data, a commitment to accountability, as well as robust Congressional oversight."
In particular, the report highlighted GAO’s repeated call for a permanent data analytics centre of excellence as one of nine recommendations left unanswered from a report it published back in March 2022.
Do Not Pay
The 2022 report on improving accountability for emergency relief funds said the US lacked a “permanent, government-wide analytic capability” to identify fraud, leaving federal agencies and oversight committees with “limited resources… to ensure robust financial stewardship.”
While one data analytics tool developed to interrogate pandemic-related spending partially addressed this need by combining more than 60 data sources, GAO said it was not permanent or government-wide.
Other recommendations included amendments to the 2014 DATA Act to re-introduce regular reviews of agency data submissions, and to clarify the responsibilities of the Treasury and Office of Management and Budget on publishing such data.
Only one recommendation, to make permanent a pilot data-sharing programme between the Social Security Administration and the Treasury’s Do Not Pay system has been acted on, GAO said this week.
The Do Not Pay platform was set up in 2011 with a similar aim to GAO’s proposed centre. It runs analytics on 23 data sources to highlight high-risk payments before they’re made and recovered a record $11.7 billion in 2025, but is only used by agencies on a voluntary basis.
Could AI stop the bleed?
In January, GAO also told Congress the use of AI tools alongside a data analytics programme could “enhance efforts” to limit improper payments, but said “mission-critical gaps” in the federal workforce’s science, technology, engineering and math skills limited its ability to deliver on this promise.
Strict data security rules, a need for more high-quality data, and the lack of a solid risk model were also keeping agencies from using generative AI for this purpose.
A human in the loop approach and the use of “reliable ground truth” data would help address these hurdles, auditors said.
The link has been copied!
