Federal Agencies Get New Guidelines for Buying and Deploying AI

A new set of government-wide guidelines released this week lays out how federal agencies should purchase and deploy artificial intelligence, requiring documented risk assessments and human oversight for decisions that affect people's rights or benefits.

The framework asks agencies to inventory the AI tools they already use, publish plain-language descriptions of what each system does, and test high-impact tools for accuracy and bias before deployment. Vendors selling to the government will need to disclose how their models were trained and evaluated.

Human in the loop

For consequential decisions, such as flagging a tax return for review or screening a benefits application, the guidelines require a meaningful human check before any adverse action. Agencies must also provide a way for the public to contest automated determinations.

Technology groups broadly welcomed the clarity, saying consistent rules make it easier to sell into the federal market. Civil liberties organizations praised the transparency requirements while urging stronger enforcement teeth.

Agencies have a year to bring existing systems into compliance and must report progress to oversight offices. The guidance stops short of banning any category of tool, instead betting that disclosure and testing will catch the worst problems early.