[ad_1]
(Bloomberg) — The US federal government’s central human resources office will begin deploying artificial intelligence to clean up its personnel records this year, marking the boldest public move yet into using the technology on sensitive records.
The Office of Personnel Management’s disclosure — in a privacy impact memo posted Wednesday — follows government-wide guidance last week directing agencies to “lean forward” on the use of the technology to improve services and cut costs.
President Donald Trump signed an executive order in January revoking former President Joe Biden’s more cautious AI policy, which Trump said put in place “barriers to American AI innovation.”
Still the White House has been reluctant to acknowledge the government’s use of AI on sensitive federal data. US agency employees and others have brought lawsuits over granting Elon Musk’s Department of Government Efficiency access to that data, in part because of privacy and civil rights concerns about the use of AI.
Musk’s DOGE teams have scattered across the government, tapping into agency databases to direct spending cuts and recommend ways to downsize the federal workforce.
White House Press Secretary Karoline Leavitt confirmed the OPM project Wednesday. The White House said the project, first reported by Fox News, would “usher in historic efficiency” at the federal personnel office.
OPM said the project would use AI to extract data from electronic personnel records, with the goal of correcting inaccurate information. OPM said the work would be done in controlled environments by approved officials and contractors, and that it would use “stateless models” for the project — meaning that the records would not be used to train or improve AI models.
Modernizing personnel records has been a top target of Musk’s DOGE teams, which Trump has directed to use technology to update federal information technology systems. OPM currently processes retirement papers by hand in a decommissioned limestone mine in Pennsylvania.
More stories like this are available on bloomberg.com
[ad_2]
